20191120.md

ArXiv cs.CV --Wed, 20 Nov 2019

1.Defective Convolutional Layers Learn Robust CNNs ⬇️

Robustness of convolutional neural networks has recently been highlighted by the adversarial examples, i.e., inputs added with well-designed perturbations which are imperceptible to humans but can cause the network to give incorrect outputs. Recent research suggests that the noises in adversarial examples break the textural structure, which eventually leads to wrong predictions by convolutional neural networks. To help a convolutional neural network make predictions relying less on textural information, we propose defective convolutional layers which contain defective neurons whose activations are set to be a constant function. As the defective neurons contain no information and are far different from the standard neurons in its spatial neighborhood, the textural features cannot be accurately extracted and the model has to seek for other features for classification, such as the shape. We first show that predictions made by the defective CNN are less dependent on textural information, but more on shape information, and further find that adversarial examples generated by the defective CNN appear to have semantic shapes. Experimental results demonstrate the defective CNN has higher defense ability than the standard CNN against various types of attack. In particular, it achieves state-of-the-art performance against transfer-based attacks without applying any adversarial training.

2.A Boost Strategy to the Generative Error Based Video Anomaly Detection Algorithms ⬇️

The generation error (GE) based algorithms show excellent performances in the task of video anomaly detection. However, in the step of anomaly detection, they have two problems: (1) Abnormal events usually occur in local areas. It reduces the saliencies of the abnormal events to use the frame-level GE as the anomaly-score. (2) Every discrimination has both advantages and disadvantages, it is difficult to aggregate multiple discriminations effectively. To address these problems, we propose a promotion strategy which is consisted of two modules. Firstly, we replace the frame-level GE with the maximum of the block-level GEs in a frame as the anomaly score. Secondly, assuming that the stricter the anomaly threshold the more reliable the anomaly detected, we propose a reliable-anomaly (R-anomaly) based multiple discriminations aggregation method. In this method, we set a strict anomaly detection threshold (SADT) for each auxiliary discrimination to detect R-anomalies. Then we use the detected R-anomalies to enhance their anomaly scores in the main discrimination. Experiments are carried out on UCSD and CUHK Avenue datasets. The results demonstrate the effectiveness of the proposed strategy and achieve state-of-the-art performance.

3.KISS: Keeping It Simple for Scene Text Recognition ⬇️

Over the past few years, several new methods for scene text recognition have been proposed. Most of these methods propose novel building blocks for neural networks. These novel building blocks are specially tailored for the task of scene text recognition and can thus hardly be used in any other tasks. In this paper, we introduce a new model for scene text recognition that only consists of off-the-shelf building blocks for neural networks. Our model (KISS) consists of two ResNet based feature extractors, a spatial transformer, and a transformer. We train our model only on publicly available, synthetic training data and evaluate it on a range of scene text recognition benchmarks, where we reach state-of-the-art or competitive performance, although our model does not use methods like 2D-attention, or image rectification.

4.Solar Event Tracking with Deep Regression Networks: A Proof of Concept Evaluation ⬇️

With the advent of deep learning for computer vision tasks, the need for accurately labeled data in large volumes is vital for any application. The increasingly available large amounts of solar image data generated by the Solar Dynamic Observatory (SDO) mission make this domain particularly interesting for the development and testing of deep learning systems. The currently available labeled solar data is generated by the SDO mission's Feature Finding Team's (FFT) specialized detection modules. The major drawback of these modules is that detection and labeling is performed with a cadence of every 4 to 12 hours, depending on the module. Since SDO image data products are created every 10 seconds, there is a considerable gap between labeled observations and the continuous data stream. In order to address this shortcoming, we trained a deep regression network to track the movement of two solar phenomena: Active Region and Coronal Hole events. To the best of our knowledge, this is the first attempt of solar event tracking using a deep learning approach. Since it is impossible to fully evaluate the performance of the suggested event tracks with the original data (only partial ground truth is available), we demonstrate with several metrics the effectiveness of our approach. With the purpose of generating continuously labeled solar image data, we present this feasibility analysis showing the great promise of deep regression networks for this task.

5.FollowMeUp Sports: New Benchmark for 2D Human Keypoint Recognition ⬇️

Human pose estimation has made significant advancement in recent years. However, the existing datasets are limited in their coverage of pose variety. In this paper, we introduce a novel benchmark FollowMeUp Sports that makes an important advance in terms of specific postures, self-occlusion and class balance, a contribution that we feel is required for future development in human body models. This comprehensive dataset was collected using an established taxonomy of over 200 standard workout activities with three different shot angles. The collected videos cover a wider variety of specific workout activities than previous datasets including push-up, squat and body moving near the ground with severe self-occlusion or occluded by some sport equipment and outfits. Given these rich images, we perform a detailed analysis of the leading human pose estimation approaches gaining insights for the success and failures of these methods.

6.Single-Stage 6D Object Pose Estimation ⬇️

Most recent 6D pose estimation frameworks first rely on a deep network to establish correspondences between 3D object keypoints and 2D image locations and then use a variant of a RANSAC-based Perspective-n-Point (PnP) algorithm. This two-stage process, however, is suboptimal: First, it is not end-to-end trainable. Second, training the deep network relies on a surrogate loss that does not directly reflect the final 6D pose estimation task.
In this work, we introduce a deep architecture that directly regresses 6D poses from correspondences. It takes as input a group of candidate correspondences for each 3D keypoint and accounts for the fact that the order of the correspondences within each group is irrelevant, while the order of the groups, that is, of the 3D keypoints, is fixed. Our architecture is generic and can thus be exploited in conjunction with existing correspondence-extraction networks so as to yield single-stage 6D pose estimation frameworks. Our experiments demonstrate that these single-stage frameworks consistently outperform their two-stage counterparts in terms of both accuracy and speed.

7.Lightweight Residual Network for The Classification of Thyroid Nodules ⬇️

Ultrasound is a useful technique for diagnosing thyroid nodules. Benign and malignant nodules that automatically discriminate in the ultrasound pictures can provide diagnostic recommendations or, improve diagnostic accuracy in the absence of specialists. The main issue here is how to collect suitable features for this particular task. We suggest here a technique for extracting features from ultrasound pictures based on the Residual U-net. We attempt to introduce significant semantic characteristics to the classification. Our model gained 95% classification accuracy.

8.Learning Modulated Loss for Rotated Object Detection ⬇️

Popular rotated detection methods usually use five parameters (coordinates of the central point, width, height, and rotation angle) to describe the rotated bounding box and l1-loss as the loss function. In this paper, we argue that the aforementioned integration can cause training instability and performance degeneration, due to the loss discontinuity resulted from the inherent periodicity of angles and the associated sudden exchange of width and height. This problem is further pronounced given the regression inconsistency among five parameters with different measurement units. We refer to the above issues as rotation sensitivity error (RSE) and propose a modulated rotation loss to dismiss the loss discontinuity. Our new loss is combined with the eight-parameter regression to further solve the problem of inconsistent parameter regression. Experiments show the state-of-art performances of our method on the public aerial image benchmark DOTA and UCAS-AOD. Its generalization abilities are also verified on ICDAR2015, HRSC2016, and FDDB. Qualitative improvements can be seen in Fig 1, and the source code will be released with the publication of the paper.

9.Distance-IoU Loss: Faster and Better Learning for Bounding Box Regression ⬇️

Bounding box regression is the crucial step in object detection. In existing methods, while $\ell_n$-norm loss is widely adopted for bounding box regression, it is not tailored to the evaluation metric, i.e., Intersection over Union (IoU). Recently, IoU loss and generalized IoU (GIoU) loss have been proposed to benefit the IoU metric, but still suffer from the problems of slow convergence and inaccurate regression. In this paper, we propose a Distance-IoU (DIoU) loss by incorporating the normalized distance between the predicted box and the target box, which converges much faster in training than IoU and GIoU losses. Furthermore, this paper summarizes three geometric factors in bounding box regression, \ie, overlap area, central point distance and aspect ratio, based on which a Complete IoU (CIoU) loss is proposed, thereby leading to faster convergence and better performance. By incorporating DIoU and CIoU losses into state-of-the-art object detection algorithms, e.g., YOLO v3, SSD and Faster RCNN, we achieve notable performance gains in terms of not only IoU metric but also GIoU metric. Moreover, DIoU can be easily adopted into non-maximum suppression (NMS) to act as the criterion, further boosting performance improvement. The source code and trained models are available at this https URL.

10.General $E(2)$-Equivariant Steerable CNNs ⬇️

The big empirical success of group equivariant networks has led in recent years to the sprouting of a great variety of equivariant network architectures. A particular focus has thereby been on rotation and reflection equivariant CNNs for planar images. Here we give a general description of $E(2)$-equivariant convolutions in the framework of Steerable CNNs. The theory of Steerable CNNs thereby yields constraints on the convolution kernels which depend on group representations describing the transformation laws of feature spaces. We show that these constraints for arbitrary group representations can be reduced to constraints under irreducible representations. A general solution of the kernel space constraint is given for arbitrary representations of the Euclidean group $E(2)$ and its subgroups. We implement a wide range of previously proposed and entirely new equivariant network architectures and extensively compare their performances. $E(2)$-steerable convolutions are further shown to yield remarkable gains on CIFAR-10, CIFAR-100 and STL-10 when used as a drop-in replacement for non-equivariant convolutions.

11.Dual affine moment invariants ⬇️

Affine transformation is one of the most common transformations in nature, which is an important issue in the field of computer vision and shape analysis. And affine transformations often occur in both shape and color space simultaneously, which can be termed as Dual-Affine Transformation (DAT). In general, we should derive invariants of different data formats separately, such as 2D color images, 3D color objects, or even higher-dimensional data. To the best of our knowledge, there is no general framework to derive invariants for all of these data formats. In this paper, we propose a general framework to derive moment invariants under DAT for objects in M-dimensional space with N channels, which can be called dual-affine moment invariants (DAMI). Following this framework, we present the generating formula of DAMI under DAT for 3D color objects. Then, we instantiated a complete set of DAMI for 3D color objects with orders and degrees no greater than 4. Finally, we analyze the characteristic of these DAMI and conduct classification experiments to evaluate the stability and discriminability of them. The results prove that DAMI is robust for DAT. Our derivation framework can be applied to data in any dimension with any number of channels.

12.Constrained R-CNN: A general image manipulation detection model ⬇️

Recently, deep learning-based models have exhibited remarkable performance for image manipulation detection. However, most of them suffer from poor universality of handcrafted or predetermined features. Meanwhile, they only focus on manipulation localization and overlook manipulation classification. To address these issues, we propose a coarse-to-fine architecture named Constrained R-CNN for complete and accurate image forensics. First, the learnable manipulation feature extractor learns a unified feature representation directly from data. Second, the attention region proposal network effectively discriminates manipulated regions for the next manipulation classification and coarse localization. Then, the skip structure fuses low-level and high-level information to refine the global manipulation features. Finally, the coarse localization information guides the model to further learn the finer local features and segment out the tampered region. Experimental results demonstrate that our model achieves state-of-the-art performance. Especially, the F1 score is increased by 28.4%, 73.2%, 13.3% on the NIST16, COVERAGE, and Columbia dataset.

13.On the Impact of Object and Sub-component Level Segmentation Strategies for Supervised Anomaly Detection within X-ray Security Imagery ⬇️

X-ray security screening is in widespread use to maintain transportation security against a wide range of potential threat profiles. Of particular interest is the recent focus on the use of automated screening approaches, including the potential anomaly detection as a methodology for concealment detection within complex electronic items. Here we address this problem considering varying segmentation strategies to enable the use of both object level and sub-component level anomaly detection via the use of secondary convolutional neural network (CNN) architectures. Relative performance is evaluated over an extensive dataset of exemplar cluttered X-ray imagery, with a focus on consumer electronics items. We find that sub-component level segmentation produces marginally superior performance in the secondary anomaly detection via classification stage, with true positive of ~98% of anomalies, with a ~3% false positive.

14.Mimic The Raw Domain: Accelerating Action Recognition in the Compressed Domain ⬇️

Video understanding usually requires expensive computation that prohibits its deployment, yet videos contain significant spatiotemporal redundancy that can be exploited. In particular, operating directly on the motion vectors and residuals in the compressed video domain can significantly accelerate the compute, by not using the raw videos which demand colossal storage capacity. Existing methods approach this task as a multiple modalities problem. In this paper we are approaching the task in a completely different way; we are looking at the data from the compressed stream as a one unit clip and propose that the residual frames can replace the original RGB frames from the raw domain. Furthermore, we are using teacher-student method to aid the network in the compressed domain to mimic the teacher network in the raw domain. We show experiments on three leading datasets (HMDB51, UCF1, and Kinetics) that approach state-of-the-art accuracy on raw video data by using compressed data. Our model MFCD-Net outperforms prior methods in the compressed domain and more importantly, our model has 11X fewer parameters and 3X fewer Flops, dramatically improving the efficiency of video recognition inference. This approach enables applying neural networks exclusively in the compressed domain without compromising accuracy while accelerating performance.

15.Weakly-Supervised Video Moment Retrieval via Semantic Completion Network ⬇️

Video moment retrieval is to search the moment that is most relevant to the given natural language query. Existing methods are mostly trained in a fully-supervised setting, which requires the full annotations of temporal boundary for each query. However, manually labeling the annotations is actually time-consuming and expensive. In this paper, we propose a novel weakly-supervised moment retrieval framework requiring only coarse video-level annotations for training. Specifically, we devise a proposal generation module that aggregates the context information to generate and score all candidate proposals in one single pass. We then devise an algorithm that considers both exploitation and exploration to select top-K proposals. Next, we build a semantic completion module to measure the semantic similarity between the selected proposals and query, compute reward and provide feedbacks to the proposal generation module for scoring refinement. Experiments on the ActivityCaptions and Charades-STA demonstrate the effectiveness of our proposed method.

16.Rethinking deep active learning: Using unlabeled data at model training ⬇️

Active learning typically focuses on training a model on few labeled examples alone, while unlabeled ones are only used for acquisition. In this work we depart from this setting by using both labeled and unlabeled data during model training across active learning cycles. We do so by using unsupervised feature learning at the beginning of the active learning pipeline and semi-supervised learning at every active learning cycle, on all available data. The former has not been investigated before in active learning, while the study of latter in the context of deep learning is scarce and recent findings are not conclusive with respect to its benefit. Our idea is orthogonal to acquisition strategies by using more data, much like ensemble methods use more models. By systematically evaluating on a number of popular acquisition strategies and datasets, we find that the use of unlabeled data during model training brings a surprising accuracy improvement in image classification, compared to the differences between acquisition strategies. We thus explore smaller label budgets, even one label per class.

17.Dense Fusion Classmate Network for Land Cover Classification ⬇️

Recently, FCNs based methods have made great progress in semantic segmentation. Different with ordinary scenes, satellite image owns specific characteristics, which elements always extend to large scope and no regular or clear boundaries. Therefore, effective mid-level structure information extremely missing, precise pixel-level classification becomes tough issues. In this paper, a Dense Fusion Classmate Network (DFCNet) is proposed to adopt in land cover classification.

18.Differentiating Features for Scene Segmentation Based on Dedicated Attention Mechanisms ⬇️

Semantic segmentation is a challenge in scene parsing. It requires both context information and rich spatial information. In this paper, we differentiate features for scene segmentation based on dedicated attention mechanisms (DF-DAM), and two attention modules are proposed to optimize the high-level and low-level features in the encoder, respectively. Specifically, we use the high-level and low-level features of ResNet as the source of context information and spatial information, respectively, and optimize them with attention fusion module and 2D position attention module, respectively. For attention fusion module, we adopt dual channel weight to selectively adjust the channel map for the highest two stage features of ResNet, and fuse them to get context information. For 2D position attention module, we use the context information obtained by attention fusion module to assist the selection of the lowest-stage features of ResNet as supplementary spatial information. Finally, the two sets of information obtained by the two modules are simply fused to obtain the prediction. We evaluate our approach on Cityscapes and PASCAL VOC 2012 datasets. In particular, there aren't complicated and redundant processing modules in our architecture, which greatly reduces the complexity, and we achieving 82.3% Mean IoU on PASCAL VOC 2012 test dataset without pre-training on MS-COCO dataset.

19.GraphTER: Unsupervised Learning of Graph Transformation Equivariant Representations via Auto-Encoding Node-wise Transformations ⬇️

Recent advances in Graph Convolutional Neural Networks (GCNNs) have shown their efficiency for non-Euclidean data on graphs, which often require a large amount of labeled data with high cost. It it thus critical to learn graph feature representations in an unsupervised manner in practice. To this end, we propose a novel unsupervised learning of Graph Transformation Equivariant Representations (GraphTER), aiming to capture intrinsic patterns of graph structure under both global and local transformations. Specifically, we allow to sample different groups of nodes from a graph and then transform them node-wise isotropically or anisotropically. Then, we self-train a representation encoder to capture the graph structures by reconstructing these node-wise transformations from the feature representations of the original and transformed graphs. In experiments, we apply the learned GraphTER to graphs of 3D point cloud data, and results on point cloud segmentation/classification show that GraphTER significantly outperforms state-of-the-art unsupervised approaches and pushes greatly closer towards the upper bound set by the fully supervised counterparts.

20.Tell Me What They're Holding: Weakly-supervised Object Detection with Transferable Knowledge from Human-object Interaction ⬇️

In this work, we introduce a novel weakly supervised object detection (WSOD) paradigm to detect objects belonging to rare classes that have not many examples using transferable knowledge from human-object interactions (HOI). While WSOD shows lower performance than full supervision, we mainly focus on HOI as the main context which can strongly supervise complex semantics in images. Therefore, we propose a novel module called RRPN (relational region proposal network) which outputs an object-localizing attention map only with human poses and action verbs. In the source domain, we fully train an object detector and the RRPN with full supervision of HOI. With transferred knowledge about localization map from the trained RRPN, a new object detector can learn unseen objects with weak verbal supervision of HOI without bounding box annotations in the target domain. Because the RRPN is designed as an add-on type, we can apply it not only to the object detection but also to other domains such as semantic segmentation. The experimental results on HICO-DET dataset show the possibility that the proposed method can be a cheap alternative for the current supervised object detection paradigm. Moreover, qualitative results demonstrate that our model can properly localize unseen objects on HICO-DET and V-COCO datasets.

21.MarioNETte: Few-shot Face Reenactment Preserving Identity of Unseen Targets ⬇️

When there is a mismatch between the target identity and the driver identity, face reenactment suffers severe degradation in the quality of the result, especially in a few-shot setting. The identity preservation problem, where the model loses the detailed information of the target leading to a defective output, is the most common failure mode. The problem has several potential sources such as the identity of the driver leaking due to the identity mismatch, or dealing with unseen large poses. To overcome such problems, we introduce components that address the mentioned problem: image attention block, target feature alignment, and landmark transformer. Through attending and warping the relevant features, the proposed architecture, called MarioNETte, produces high-quality reenactments of unseen identities in a few-shot setting. In addition, the landmark transformer dramatically alleviates the identity preservation problem by isolating the expression geometry through landmark disentanglement. Comprehensive experiments are performed to verify that the proposed framework can generate highly realistic faces, outperforming all other baselines, even under a significant mismatch of facial characteristics between the target and the driver.

22.Neural Network Pruning with Residual-Connections and Limited-Data ⬇️

Filter level pruning is an effective method to accelerate the inference speed of deep CNN models. Although numerous pruning algorithms have been proposed, there are still two open issues. The first problem is how to prune residual connections. Most previous filter level pruning algorithms only prune channels inside residual blocks, leaving the number of output channels unchanged. We show that pruning both channels inside and outside the residual connections is crucial to achieve better performance. The second issue is pruning with limited data. We observe an interesting phenomenon: directly pruning on a small dataset is usually worse than fine-tuning a small model which is pruned or trained from scratch on the large dataset. In this paper, we propose a novel method, namely Compression Using Residual-connections and Limited-data (CURL), to tackle these two challenges. Experiments on the large scale dataset demonstrate the effectiveness of CURL. CURL significantly outperforms previous state-of-the-art methods on ImageNet. More importantly, when pruning on small datasets, CURL achieves comparable or much better performance than fine-tuning a pretrained small model.

23.DebFace: De-biasing Face Recognition ⬇️

We address the problem of bias in automated face recognition algorithms, where errors are consistently lower on certain cohorts belonging to specific demographic groups. We present a novel de-biasing adversarial network that learns to extract disentangled feature representations for both unbiased face recognition and demographics estimation. The proposed network consists of one identity classifier and three demographic classifiers (for gender, age, and race) that are trained to distinguish identity and demographic attributes, respectively. Adversarial learning is adopted to minimize correlation among feature factors so as to abate bias influence from other factors. We also design a new scheme to combine demographics with identity features to strengthen robustness of face representation in different demographic groups. The experimental results show that our approach is able to reduce bias in face recognition as well as demographics estimation while achieving state-of-the-art performance.

24.Two-Stream FCNs to Balance Content and Style for Style Transfer ⬇️

Style transfer is to render given image contents in given styles, and it has an important role in both computer vision fundamental research and industrial applications. Following the success of deep learning based approaches, this problem has been re-launched very recently, but still remains a difficult task because of trade-off between preserving contents and faithful rendering of styles. In this paper, we propose an end-to-end two-stream Fully Convolutional Networks (FCNs) aiming at balancing the contributions of the content and the style in rendered images. Our proposed network consists of the encoder and decoder parts. The encoder part utilizes a FCN for content and a FCN for style where the two FCNs have feature injections and are independently trained to preserve the semantic content and to learn the faithful style representation in each. The semantic content feature and the style representation feature are then concatenated adaptively and fed into the decoder to generate style-transferred (stylized) images. In order to train our proposed network, we employ a loss network, the pre-trained VGG-16, to compute content loss and style loss, both of which are efficiently used for the feature injection as well as the feature concatenation. Our intensive experiments show that our proposed model generates more balanced stylized images in content and style than state-of-the-art methods. Moreover, our proposed network achieves efficiency in speed.

25.IFQ-Net: Integrated Fixed-point Quantization Networks for Embedded Vision ⬇️

Deploying deep models on embedded devices has been a challenging problem since the great success of deep learning based networks. Fixed-point networks, which represent their data with low bits fixed-point and thus give remarkable savings on memory usage, are generally preferred. Even though current fixed-point networks employ relative low bits (e.g. 8-bits), the memory saving is far from enough for the embedded devices. On the other hand, quantization deep networks, for example XNOR-Net and HWGQNet, quantize the data into 1 or 2 bits resulting in more significant memory savings but still contain lots of floatingpoint data. In this paper, we propose a fixed-point network for embedded vision tasks through converting the floatingpoint data in a quantization network into fixed-point. Furthermore, to overcome the data loss caused by the conversion, we propose to compose floating-point data operations across multiple layers (e.g. convolution, batch normalization and quantization layers) and convert them into fixedpoint. We name the fixed-point network obtained through such integrated conversion as Integrated Fixed-point Quantization Networks (IFQ-Net). We demonstrate that our IFQNet gives 2.16x and 18x more savings on model size and runtime feature map memory respectively with similar accuracy on ImageNet. Furthermore, based on YOLOv2, we design IFQ-Tinier-YOLO face detector which is a fixed-point network with 256x reduction in model size (246k Bytes) than Tiny-YOLO. We illustrate the promising performance of our face detector in terms of detection rate on Face Detection Data Set and Bencmark (FDDB) and qualitative results of detecting small faces of Wider Face dataset.

26.A novel method for identifying the deep neural network model with the Serial Number ⬇️

Deep neural network (DNN) with the state of art performance has emerged as a viable and lucrative business service. However, those impressive performances require a large number of computational resources, which comes at a high cost for the model creators. The necessity for protecting DNN models from illegal reproducing and distribution appears salient now. Recently, trigger-set watermarking, breaking the white-box restriction, relying on adversarial training pre-defined (incorrect) labels for crafted inputs, and subsequently using them to verify the model authenticity, has been the main topic of DNN ownership verification. While these methods have successfully demonstrated robustness against removal attacks, few are effective against the tampering attacks from competitors forging the fake watermarks and dogging in the manager. In this paper, we put forth a new framework of the trigger-set watermark by embedding a unique Serial Number (relatedness less original labels) to the deep neural network for model ownership identification, which is both robust to model pruning and resist to tampering attacks. Experiment results demonstrate that the DNN Serial Number only incurs slight accuracy degradation of the original performance and is valid for ownership verification.

27.Poison as a Cure: Detecting & Neutralizing Variable-Sized Backdoor Attacks in Deep Neural Networks ⬇️

Deep learning models have recently shown to be vulnerable to backdoor poisoning, an insidious attack where the victim model predicts clean images correctly but classifies the same images as the target class when a trigger poison pattern is added. This poison pattern can be embedded in the training dataset by the adversary. Existing defenses are effective under certain conditions such as a small size of the poison pattern, knowledge about the ratio of poisoned training samples or when a validated clean dataset is available. Since a defender may not have such prior knowledge or resources, we propose a defense against backdoor poisoning that is effective even when those prerequisites are not met. It is made up of several parts: one to extract a backdoor poison signal, detect poison target and base classes, and filter out poisoned from clean samples with proven guarantees. The final part of our defense involves retraining the poisoned model on a dataset augmented with the extracted poison signal and corrective relabeling of poisoned samples to neutralize the backdoor. Our approach has shown to be effective in defending against backdoor attacks that use both small and large-sized poison patterns on nine different target-base class pairs from the CIFAR10 dataset.

28.Reliability Does Matter: An End-to-End Weakly Supervised Semantic Segmentation Approach ⬇️

Weakly supervised semantic segmentation is a challenging task as it only takes image-level information as supervision for training but produces pixel-level predictions for testing. To address such a challenging task, most recent state-of-the-art approaches propose to adopt two-step solutions, \emph{i.e. } 1) learn to generate pseudo pixel-level masks, and 2) engage FCNs to train the semantic segmentation networks with the pseudo masks. However, the two-step solutions usually employ many bells and whistles in producing high-quality pseudo masks, making this kind of methods complicated and inelegant. In this work, we harness the image-level labels to produce reliable pixel-level annotations and design a fully end-to-end network to learn to predict segmentation maps. Concretely, we firstly leverage an image classification branch to generate class activation maps for the annotated categories, which are further pruned into confident yet tiny object/background regions. Such reliable regions are then directly served as ground-truth labels for the parallel segmentation branch, where a newly designed dense energy loss function is adopted for optimization. Despite its apparent simplicity, our one-step solution achieves competitive mIoU scores (\emph{val}: 62.6, \emph{test}: 62.9) on Pascal VOC compared with those two-step state-of-the-arts. By extending our one-step method to two-step, we get a new state-of-the-art performance on the Pascal VOC (\emph{val}: 66.3, \emph{test}: 66.5).

29.Simultaneous Region Localization and Hash Coding for Fine-grained Image Retrieval ⬇️

Fine-grained image hashing is a challenging problem due to the difficulties of discriminative region localization and hash code generation. Most existing deep hashing approaches solve the two tasks independently. While these two tasks are correlated and can reinforce each other. In this paper, we propose a deep fine-grained hashing to simultaneously localize the discriminative regions and generate the efficient binary codes. The proposed approach consists of a region localization module and a hash coding module. The region localization module aims to provide informative regions to the hash coding module. The hash coding module aims to generate effective binary codes and give feedback for learning better localizer. Moreover, to better capture subtle differences, multi-scale regions at different layers are learned without the need of bounding-box/part annotations. Extensive experiments are conducted on two public benchmark fine-grained datasets. The results demonstrate significant improvements in the performance of our method relative to other fine-grained hashing algorithms.

30.Towards a complete 3D morphable model of the human head ⬇️

Three-dimensional Morphable Models (3DMMs) are powerful statistical tools for representing the 3D shapes and textures of an object class. Here we present the most complete 3DMM of the human head to date that includes face, cranium, ears, eyes, teeth and tongue. To achieve this, we propose two methods for combining existing 3DMMs of different overlapping head parts: i. use a regressor to complete missing parts of one model using the other, ii. use the Gaussian Process framework to blend covariance matrices from multiple models. Thus we build a new combined face-and-head shape model that blends the variability and facial detail of an existing face model (the LSFM) with the full head modelling capability of an existing head model (the LYHM). Then we construct and fuse a highly-detailed ear model to extend the variation of the ear shape. Eye and eye region models are incorporated into the head model, along with basic models of the teeth, tongue and inner mouth cavity. The new model achieves state-of-the-art performance. We use our model to reconstruct full head representations from single, unconstrained images allowing us to parameterize craniofacial shape and texture, along with the ear shape, eye gaze and eye color.

31.Streetify: Using Street View Imagery And Deep Learning For Urban Streets Development ⬇️

The classification of streets on road networks has been focused on the vehicular transportational features of streets such as arterials, major roads, minor roads and so forth based on their transportational use. City authorities on the other hand have been shifting to more urban inclusive planning of streets, encompassing the side use of a street combined with the transportational features of a street. In such classification schemes, streets are labeled for example as commercial throughway, residential neighborhood, park etc. This modern approach to urban planning has been adopted by major cities such as the city of San Francisco, the states of Florida and Pennsylvania among many others. Currently, the process of labeling streets according to their contexts is manual and hence is tedious and time consuming. In this paper, we propose an approach to collect and label imagery data then deploy advancements in computer vision towards modern urban planning. We collect and label street imagery then train deep convolutional neural networks (CNN) to perform the classification of street context. We show that CNN models can perform well achieving accuracies in the 81% to 87%, we then visualize samples from the embedding space of streets using the t-SNE method and apply class activation mapping methods to interpret the features in street imagery contributing to output classification from a model.

32.Segmentation Guided Attention Network for Crowd Counting via Curriculum Learning ⬇️

Crowd counting using deep convolutional neural networks (CNN) has achieved encouraging progress in the last couple of years. Novel network architectures have been designed to handle the scale variance issue in crowd images. For this purpose, the ideas of using multi-column networks with different convolution kernel sizes and rich feature fusion have been prevalent in literature. Recent works have shown the effectiveness of \textit{Inception} modules in crowd counting due to its ability to capture multi-scale visual information via the fusion of features from multi-column networks. However, the existing crowd counting networks built with \textit{Inception} modules usually have a small number of layers and only employ the basic type of \textit{Inception} modules. In this paper, we investigate the use of pre-trained \textit{Inception} model for crowd counting. Specifically, we firstly benchmark the baseline \textit{Inception-v3} models on commonly used crowd counting datasets and show its superiority to other existing models. Subsequently, we present a Segmentation Guided Attention Network (SGANet) with the \textit{Inception-v3} as the backbone for crowd counting. We also propose a novel curriculum learning strategy for more efficient training of crowd counting networks. Finally, we conduct thorough experiments to compare the performance of SGANet and other state-of-the-art models. The experimental results validate the effectiveness of the segmentation guided attention layer and the curriculum learning strategy in crowd counting.

33.Simultaneous Mapping and Target Driven Navigation ⬇️

This work presents a modular architecture for simultaneous mapping and target driven navigation in indoors environments. The semantic and appearance stored in 2.5D map is distilled from RGB images, semantic segmentation and outputs of object detectors by convolutional neural networks. Given this representation, the mapping module learns to localize the agent and register consecutive observations in the map. The navigation task is then formulated as a problem of learning a policy for reaching semantic targets using current observations and the up-to-date map. We demonstrate that the use of semantic information improves localization accuracy and the ability of storing spatial semantic map aids the target driven navigation policy. The two modules are evaluated separately and jointly on Active Vision Dataset and Matterport3D environments, demonstrating improved performance on both localization and navigation tasks.

34.Improving the Robustness of Capsule Networks to Image Affine Transformations ⬇️

Convolutional neural networks (CNNs) achieve translational invariance using pooling operations, which do not maintain the spatial relationship in the learned representations. Hence, they cannot extrapolate their understanding of the geometric transformation of inputs. Recently, Capsule Networks (CapsNets) have been proposed to tackle this problem. In CapsNets, each entity is represented by a vector and routed to high-level entities by a dynamic routing algorithm. The CapsNets have been shown to be more robust than CNNs to affine transformations of inputs. However, there is still a huge gap between their performance on transformed inputs compared to untransformed versions. In this work, we first revisit the routing procedure by (un)rolling its forward and backward passes. Our investigation reveals that the routing procedure contributes neither to generalization ability nor to the affine robustness of the CapsNets.
Furthermore, we explore the limitations of capsule transformations and propose affine CapsNets (Aff-CapsNets) that are more robust to affine transformations. On our benchmark task where models are trained on the MNIST dataset and tested on the AffNIST dataset, our Aff-CapsNets improve the benchmark performance by a large margin (from 79% to 93.21%), without using a routing mechanism. We also demonstrate the superiority of Aff-CapsNets on a real-world Brain Tumor Type classification task.

35.TracKlinic: Diagnosis of Challenge Factors in Visual Tracking ⬇️

Generic visual tracking is difficult due to many challenge factors (e.g., occlusion, blur, etc.). Each of these factors may cause serious problems for a tracking algorithm, and when they work together can make things even more complicated. Despite a great amount of efforts devoted to understanding the behavior of tracking algorithms, reliable and quantifiable ways for studying the per factor tracking behavior remain barely available. Addressing this issue, in this paper we contribute to the community a tracking diagnosis toolkit, TracKlinic, for diagnosis of challenge factors of tracking algorithms.
TracKlinic consists of two novel components focusing on the data and analysis aspects, respectively. For the data component, we carefully prepare a set of 2,390 annotated videos, each involving one and only one major challenge factor. When analyzing an algorithm for a specific challenge factor, such one-factor-per-sequence rule greatly inhibits the disturbance from other factors and consequently leads to more faithful analysis. For the analysis component, given the tracking results on all sequences, it investigates the behavior of the tracker under each individual factor and generates the report automatically. With TracKlinic, a thorough study is conducted on ten state-of-the-art trackers on nine challenge factors (including two compound ones). The results suggest that, heavy shape variation and occlusion are the two most challenging factors faced by most trackers. Besides, out-of-view, though does not happen frequently, is often fatal. By sharing TracKlinic, we expect to make it much easier for diagnosing tracking algorithms, and to thus facilitate developing better ones.

36.Neighborhood Watch: Representation Learning with Local-Margin Triplet Loss and Sampling Strategy for K-Nearest-Neighbor Image Classification ⬇️

Deep representation learning using triplet network for classification suffers from a lack of theoretical foundation and difficulty in tuning both the network and classifiers for performance. To address the problem, local-margin triplet loss along with local positive and negative mining strategy is proposed with theory on how the strategy integrate nearest-neighbor hyper-parameter with triplet learning to increase subsequent classification performance. Results in experiments with 2 public datasets, MNIST and Cifar-10, and 2 small medical image datasets demonstrate that proposed strategy outperforms end-to-end softmax and typical triplet loss in settings without data augmentation while maintaining utility of transferable feature for related tasks. The method serves as a good performance baseline where end-to-end methods encounter difficulties such as small sample data with limited allowable data augmentation.

37.Towards Good Practices for Instance Segmentation ⬇️

Instance Segmentation is an interesting yet challenging task in computer vision. In this paper, we conduct a series of refinements with the Hybrid Task Cascade (HTC) Network, and empirically evaluate their impact on the final model performance through ablation studies. By taking all the refinements, we achieve 0.47 on the COCO test-dev dataset and 0.47 on the COCO test-challenge dataset.

38.Towards Good Practices for Multi-Person Pose Estimation ⬇️

Multi-Person Pose Estimation is an interesting yet challenging task in computer vision. In this paper, we conduct a series of refinements with the MSPN and PoseFix Networks, and empirically evaluate their impact on the final model performance through ablation studies. By taking all the refinements, we achieve 78.7 on the COCO test-dev dataset and 76.3 on the COCO test-challenge dataset.

39.Inverse Graphics: Unsupervised Learning of 3D Shapes from Single Images ⬇️

Using generative models for Inverse Graphics is an active area of research. However, most works focus on developing models for supervised and semi-supervised methods. In this paper, we study the problem of unsupervised learning of 3D geometry from single images. Our approach is to use a generative model that produces 2-D images as projections of a latent 3D voxel grid, which we train either as a variational auto-encoder or using adversarial methods. Our contributions are as follows: First, we show how to recover 3D shape and pose from general datasets such as MNIST, and MNIST Fashion in good quality. Second, we compare the shapes learned using adversarial and variational methods. Adversarial approach gives denser 3D shapes. Third, we explore the idea of modelling the pose of an object as uniform distribution to recover 3D shape from a single image. Our experiment with the CelebA dataset \cite{liu2015faceattributes} proves that we can recover complete 3D shape from a single image when the object is symmetric along one, or more axis whilst results obtained using ModelNet40 \cite{wu20153d} show the potential side-effects, in which the model learns 3D shapes such that it can render the same image from any viewpoint. Forth, we present a general end-to-end approach to learning 3D shapes from single images in a completely unsupervised fashion by modelling the factors of variation such as azimuth as independent latent variables. Our method makes no assumptions about the dataset, and can work with synthetic as well as real images (i.e. unsupervised in true sense). We present our results, by training the model using the $\mu$-VAE objective \cite{ucar2019bridging} and a dataset combining all images from MNIST, MNIST Fashion, CelebA and six categories of ModelNet40. The model is able to learn 3D shapes and the pose in qood quality and leverages information learned across all datasets.

40.Privacy Preserving Gaze Estimation using Synthetic Images via a Randomized Encoding Based Framework ⬇️

Eye tracking is handled as one of the key technologies for applications which assess and evaluate human attention, behavior and biometrics, especially using gaze, pupillary and blink behaviors. One of the main challenges with regard to the social acceptance of eye-tracking technology is however the preserving of sensitive and personal information. To tackle this challenge, we employed a privacy-preserving framework based on randomized encoding to train a Support Vector Regression model on synthetic eye images privately to estimate human gaze. During the computation, none of the parties learns about the data or the result that any other party has. Furthermore, the party that trains the model cannot reconstruct pupil, blink or visual scanpath. The experimental results showed that our privacy preserving framework is also capable of working in real-time, as accurate as a non-private version of it and could be extended to other eye-tracking related problems.

41.Fitness Done Right: a Real-time Intelligent Personal Trainer for Exercise Correction ⬇️

Keeping fit has been increasingly important for people nowadays. However, people may not get expected exercise results without following professional guidance while hiring personal trainers is expensive. In this paper, an effective real-time system called Fitness Done Right (FDR) is proposed for helping people exercise correctly on their own. The system includes detecting human body parts, recognizing exercise pose and detecting errors for test poses as well as giving correction advice. Generally, two branch multi-stage CNN is used for training data sets in order to learn human body parts and associations. Then, considering two poses, which are plank and squat in our model, we design a detection algorithm, combining Euclidean and angle distances, to determine the pose in the image. Finally, key values for key features of the two poses are computed correspondingly in the pose error detection part, which helps give correction advice. We conduct our system in real-time situation with error rate down to $1.2%$, and the screenshots of experimental results are also presented.

42.Training Set Affect on Super Resolution for Automated Target Recognition ⬇️

Do we need to have more expensive remote sensing satellites when we could use single image super-resolution (SISR) to get the spatial resolution that we want? By using a Super Resolution Generative Adversarial Network, (SRGAN) we can get higher resolution images. Previous work by Shermeyer et al. [1] have used SISR as a preprocessing step describe an increase in mAP of 10-36 % in object detection for native 30cm to 15cm satellite imagery. This suggests a possible improvement to automated target recognition in image classification and object detection. The SRGAN takes a low-resolution image and maps it to a high-resolution image creating the super resolution product. We train 5 SRGANs on different land-use classes (e.g. agriculture, cities) and find the qualitative and quantitative differences in SISR, binary classification, and object detection performance.

43.Dont Even Look Once: Synthesizing Features for Zero-Shot Detection ⬇️

Zero-shot detection, namely, localizing both seen and unseen objects, increasingly gains importance for large-scale applications, with large number of object classes, since, collecting sufficient annotated data with ground truth bounding boxes is simply not scalable. While vanilla deep neural networks deliver high performance for objects available during training, unseen object detection degrades significantly. At a fundamental level, while vanilla detectors are capable of proposing bounding boxes, which include unseen objects, they are often incapable of assigning high-confidence to unseen objects, due to the inherent precision/recall tradeoffs that requires rejecting background objects. We propose a novel detection algorithm Dont Even Look Once (DELO), that synthesizes visual features for unseen objects and augments existing training algorithms to incorporate unseen object detection. Our proposed scheme is evaluated on Pascal VOC and MSCOCO, and we demonstrate significant improvements in test accuracy over vanilla and other state-of-art zero-shot detectors

44.Copy-Move Forgery Classification via Unsupervised Domain Adaptation ⬇️

In the current era, image manipulation is becoming increasingly easier, yielding more natural looking images, owing to the modern tools in image processing and computer vision techniques. The task of the segregation of forged images has become very challenging. To tackle such problems, publicly available datasets are insufficient. In this paper, we propose to create a synthetic forged dataset using deep semantic image inpainting algorithm. Furthermore, we use an unsupervised domain adaptation network to detect copy-move forgery in images. Our approach can be helpful in those cases, where the classification of data is unavailable.

45.CAGFuzz: Coverage-Guided Adversarial Generative Fuzzing Testing of Deep Learning Systems ⬇️

Deep Learning systems (DL) based on Deep Neural Networks (DNNs) are more and more used in various aspects of our life, including unmanned vehicles, speech processing, and robotics. However, due to the limited dataset and the dependence on manual labeling data, DNNs often fail to detect their erroneous behaviors, which may lead to serious problems. Several approaches have been proposed to enhance the input examples for testing DL systems. However, they have the following limitations. First, they design and generate adversarial examples from the perspective of model, which may cause low generalization ability when they are applied to other models. Second, they only use surface feature constraints to judge the difference between the adversarial example generated and the original example. The deep feature constraints, which contain high-level semantic information, such as image object category and scene semantics are completely neglected. To address these two problems, in this paper, we propose CAGFuzz, a Coverage-guided Adversarial Generative Fuzzing testing approach, which generates adversarial examples for a targeted DNN to discover its potential defects. First, we train an adversarial case generator (AEG) from the perspective of general data set. Second, we extract the depth features of the original and adversarial examples, and constrain the adversarial examples by cosine similarity to ensure that the semantic information of adversarial examples remains unchanged. Finally, we retrain effective adversarial examples to improve neuron testing coverage rate. Based on several popular data sets, we design a set of dedicated experiments to evaluate CAGFuzz. The experimental results show that CAGFuzz can improve the neuron coverage rate, detect hidden errors, and also improve the accuracy of the target DNN.

46.BiNet: Degraded-Manuscript Binarization in Diverse Document Textures and Layouts using Deep Encoder-Decoder Networks ⬇️

Handwritten document-image binarization is a semantic segmentation process to differentiate ink pixels from background pixels. It is one of the essential steps towards character recognition, writer identification, and script-style evolution analysis. The binarization task itself is challenging due to the vast diversity of writing styles, inks, and paper materials. It is even more difficult for historical manuscripts due to the aging and degradation of the documents over time. One of such manuscripts is the Dead Sea Scrolls (DSS) image collection, which poses extreme challenges for the existing binarization techniques. This article proposes a new binarization technique for the DSS images using the deep encoder-decoder networks. Although the artificial neural network proposed here is primarily designed to binarize the DSS images, it can be trained on different manuscript collections as well. Additionally, the use of transfer learning makes the network already utilizable for a wide range of handwritten documents, making it a unique multi-purpose tool for binarization. Qualitative results and several quantitative comparisons using both historical manuscripts and datasets from handwritten document image binarization competition (H-DIBCO and DIBCO) exhibit the robustness and the effectiveness of the system. The best performing network architecture proposed here is a variant of the U-Net encoder-decoders.

47.A Smartphone-Based Skin Disease Classification Using MobileNet CNN ⬇️

The MobileNet model was used by applying transfer learning on the 7 skin diseases to create a skin disease classification system on Android application. The proponents gathered a total of 3,406 images and it is considered as imbalanced dataset because of the unequal number of images on its classes. Using different sampling method and preprocessing of input data was explored to further improved the accuracy of the MobileNet. Using under-sampling method and the default preprocessing of input data achieved an 84.28% accuracy. While, using imbalanced dataset and default preprocessing of input data achieved a 93.6% accuracy. Then, researchers explored oversampling the dataset and the model attained a 91.8% accuracy. Lastly, by using oversampling technique and data augmentation on preprocessing the input data provide a 94.4% accuracy and this model was deployed on the developed Android application.

48.Visual Dialogue State Tracking for Question Generation ⬇️

GuessWhat?! is a visual dialogue task between a guesser and an oracle. The guesser aims to locate an object supposed by the oracle oneself in an image by asking a sequence of Yes/No questions. Asking proper questions with the progress of dialogue is vital for achieving successful final guess. As a result, the progress of dialogue should be properly represented and tracked. Previous models for question generation pay less attention on the representation and tracking of dialogue states, and therefore are prone to asking low quality questions such as repeated questions. This paper proposes visual dialogue state tracking (VDST) based method for question generation. A visual dialogue state is defined as the distribution on objects in the image as well as representations of objects. Representations of objects are updated with the change of the distribution on objects. An object-difference based attention is used to decode new question. The distribution on objects is updated by comparing the question-answer pair and objects. Experimental results on GuessWhat?! dataset show that our model significantly outperforms existing methods and achieves new state-of-the-art performance. It is also noticeable that our model reduces the rate of repeated questions from more than 50% to 18.85% compared with previous state-of-the-art methods.

49.VITON-GAN: Virtual Try-on Image Generator Trained with Adversarial Loss ⬇️

Generating a virtual try-on image from in-shop clothing images and a model person's snapshot is a challenging task because the human body and clothes have high flexibility in their shapes. In this paper, we develop a Virtual Try-on Generative Adversarial Network (VITON-GAN), that generates virtual try-on images using images of in-shop clothing and a model person. This method enhances the quality of the generated image when occlusion is present in a model person's image (e.g., arms crossed in front of the clothes) by adding an adversarial mechanism in the training pipeline.

50.WaveletKernelNet: An Interpretable Deep Neural Network for Industrial Intelligent Diagnosis ⬇️

Convolutional neural network (CNN), with ability of feature learning and nonlinear mapping, has demonstrated its effectiveness in prognostics and health management (PHM). However, explanation on the physical meaning of a CNN architecture has rarely been studied. In this paper, a novel wavelet driven deep neural network termed as WaveletKernelNet (WKN) is presented, where a continuous wavelet convolutional (CWConv) layer is designed to replace the first convolutional layer of the standard CNN. This enables the first CWConv layer to discover more meaningful filters. Furthermore, only the scale parameter and translation parameter are directly learned from raw data at this CWConv layer. This provides a very effective way to obtain a customized filter bank, specifically tuned for extracting defect-related impact component embedded in the vibration signal. In addition, three experimental verification using data from laboratory environment are carried out to verify effectiveness of the proposed method for mechanical fault diagnosis. The results show the importance of the designed CWConv layer and the output of CWConv layer is interpretable. Besides, it is found that WKN has fewer parameters, higher fault classification accuracy and faster convergence speed than standard CNN.

51.Logo-2K+: A Large-Scale Logo Dataset for Scalable Logo Classification ⬇️

Logo classification has gained increasing attention for its various applications, such as copyright infringement detection, product recommendation and contextual advertising. Compared with other types of object images, the real-world logo images have larger variety in logo appearance and more complexity in their background. Therefore, recognizing the logo from images is challenging. To support efforts towards scalable logo classification task, we have curated a dataset, Logo-2K+, a new large-scale publicly available real-world logo dataset with 2,341 categories and 167,140 images. Compared with existing popular logo datasets, such as FlickrLogos-32 and LOGO-Net, Logo-2K+ has more comprehensive coverage of logo categories and larger quantity of logo images. Moreover, we propose a Discriminative Region Navigation and Augmentation Network (DRNA-Net), which is capable of discovering more informative logo regions and augmenting these image regions for logo classification. DRNA-Net consists of four sub-networks: the navigator sub-network first selected informative logo-relevant regions guided by the teacher sub-network, which can evaluate its confidence belonging to the ground-truth logo class. The data augmentation sub-network then augments the selected regions via both region cropping and region dropping. Finally, the scrutinizer sub-network fuses features from augmented regions and the whole image for logo classification. Comprehensive experiments on Logo-2K+ and other three existing benchmark datasets demonstrate the effectiveness of proposed method. Logo-2K+ and the proposed strong baseline DRNA-Net are expected to further the development of scalable logo image recognition, and the Logo-2K+ dataset can be found at this https URL.

52.Cluster-wise Unsupervised Hashing for Cross-Modal Similarity Search ⬇️

In this paper, we present a new cluster-wise unsupervised hashing (CUH) approach to learn compact binary codes for cross-modal similarity retrieval. We develop a discrete optimization method to jointly learn binary codes and the corresponding hash functions for each modality which can improve the performance, unlike existing cross-modal hashing methods that often drop the binary constraints to obtain the binary codes. Moreover, considering the semantic consistency between observed modalities, our CUH generates one unified hash code for all observed modalities of any instance. Specifically, we construct a co-training framework for learning to hash, in which we simultaneously realize the multi-view clustering and the learning of hash. Firstly, our CUH utilize the re-weighted discriminatively embedded K-means for multi-view clustering to learn the corresponding dimension reduced data and the cluster centroid points in the low-dimensional common subspaces, which are used as the approximation to the corresponding hash codes of original data and the cluster-wise code-prototypes respectively. Secondly, in the process for learning of hash, these cluster-wise code-prototypes can guide the learning of the codes to further improve the performance of the binary codes. The reasonableness and effectiveness of CUH is well demonstrated by comprehensive experiments on diverse benchmark datasets.

53.Patch augmentation: Towards efficient decision boundaries for neural networks ⬇️

In this paper we propose a new augmentation technique, called patch augmentation, that, in our experiments, improves model accuracy and makes networks more robust to adversarial attacks. In brief, this data-independent approach creates new image data based on image/label pairs, where a patch from one of the two images in the pair is superimposed on to the other image, creating a new augmented sample. The new image's label is a linear combination of the image pair's corresponding labels. Initial experiments show a several percentage point increase in accuracy on CIFAR-10, from a baseline of 80.6% to 86.8%. CIFAR-100 sees better improvements still. Networks trained using patch augmentation are also more robust to adversarial attacks, which we demonstrate using the Fast Gradient Sign Method. An adversarial misclassification, or adversarial attack, occurs when an image that should seemingly be easily classified correctly by the network is suddenly classified as belonging to a completely different class-and with high confidence. Such occurrences are difficult to diagnose and are a cause of much concern in artificial intelligence research, as any model trained with empirical risk minimisation seems to be vulnerable to such attacks. The ease at which neural networks are susceptible to adversarial perturbations are partially the result of images lying close to the decision boundaries that are typically learned by neural networks during their training. Patch augmentation is an attempt to train more efficient decision boundaries.

54.ASV: Accelerated Stereo Vision System ⬇️

Estimating depth from stereo vision cameras, i.e., "depth from stereo", is critical to emerging intelligent applications deployed in energy- and performance-constrained devices, such as augmented reality headsets and mobile autonomous robots. While existing stereo vision systems make trade-offs between accuracy, performance and energy-efficiency, we describe ASV, an accelerated stereo vision system that simultaneously improves both performance and energy-efficiency while achieving high accuracy. The key to ASV is to exploit unique characteristics inherent to stereo vision, and apply stereo-specific optimizations, both algorithmically and computationally. We make two contributions. Firstly, we propose a new stereo algorithm, invariant-based stereo matching (ISM), that achieves significant speedup while retaining high accuracy. The algorithm combines classic "hand-crafted" stereo algorithms with recent developments in Deep Neural Networks (DNNs), by leveraging the correspondence invariant unique to stereo vision systems. Secondly, we observe that the bottleneck of the ISM algorithm is the DNN inference, and in particular the deconvolution operations that introduce massive compute-inefficiencies. We propose a set of software optimizations that mitigate these inefficiencies. We show that with less than 0.5% hardware area overhead, these algorithmic and computational optimizations can be effectively integrated within a conventional DNN accelerator. Overall, ASV achieves 5x speedup and 85% energy saving with 0.02% accuracy loss compared to today DNN-based stereo vision systems.

55.Cross-modal supervised learning for better acoustic representations ⬇️

Obtaining large-scale human-labeled datasets to train acoustic representation models is a very challenging task. On the contrary, we can easily collect data with machine-generated labels. In this work, we propose to exploit machine-generated labels to learn better acoustic representations, based on the synchronization between vision and audio. Firstly, we collect a large-scale video dataset with 15 million samples, which totally last 16,320 hours. Each video is 3 to 5 seconds in length and annotated automatically by publicly available visual and audio classification models. Secondly, we train various classical convolutional neural networks (CNNs) including VGGish, ResNet_50 and Mobilenet_v2. We also make several improvements to VGGish and achieve better results. Finally, we transfer our models on three external standard benchmarks for audio classification task, and achieve significant performance boost over the state-of-the-art results. Models and codes will be released in the future.

56.Face shape classification using Inception v3 ⬇️

In this paper, we present experimental results obtained from retraining the last layer of the Inception v3 model in classifying images of human faces into one of five basic face shapes. The accuracy of the retrained Inception v3 model was compared with that of the following classification methods that uses facial landmark distance ratios and angles as features: linear discriminant analysis (LDA), support vector machines with linear kernel (SVM-LIN), support vector machines with radial basis function kernel (SVM-RBF), artificial neural networks or multilayer perceptron (MLP), and k-nearest neighbors (KNN). All classifiers were trained and tested using a total of 500 images of female celebrities with known face shapes collected from the Internet. Results show that training accuracy and overall accuracy ranges from 98.0% to 100% and from 84.4% to 84.8% for Inception v3 and from 50.6% to 73.0% and from 36.4% to 64.6% for the other classifiers depending on the training set size used. This result shows that the retrained Inception v3 model was able to fit the training data well and outperform the other classifiers without the need to handpick specific features to include in model training. Future work should consider expanding the labeled dataset, preferably one that can also be freely distributed to the research community, so that proper model cross-validation can be performed. As far as we know, this is the first in the literature to use convolutional neural networks in face-shape classification. The scripts are available at this https URL.

57.Vision-Language Navigation with Self-Supervised Auxiliary Reasoning Tasks ⬇️

Vision-Language Navigation (VLN) is a task where agents learn to navigate following natural language instructions. The key to this task is to perceive both the visual scene and natural language sequentially. Conventional approaches exploit the vision and language features in cross-modal grounding. However, the VLN task remains challenging, since previous works have neglected the rich semantic information contained in the environment (such as implicit navigation graphs or sub-trajectory semantics). In this paper, we introduce Auxiliary Reasoning Navigation (AuxRN), a framework with four self-supervised auxiliary reasoning tasks to take advantage of the additional training signals derived from the semantic information. The auxiliary tasks have four reasoning objectives: explaining the previous actions, estimating the navigation progress, predicting the next orientation, and evaluating the trajectory consistency. As a result, these additional training signals help the agent to acquire knowledge of semantic representations in order to reason about its activity and build a thorough perception of the environment. Our experiments indicate that auxiliary reasoning tasks improve both the performance of the main task and the model generalizability by a large margin. Empirically, we demonstrate that an agent trained with self-supervised auxiliary reasoning tasks substantially outperforms the previous state-of-the-art method, being the best existing approach on the standard benchmark.

58.Co-Attentive Equivariant Neural Networks: Focusing Equivariance On Transformations Co-Occurring In Data ⬇️

Equivariance is a nice property to have as it produces much more parameter efficient neural architectures and preserves the structure of the input through the feature mapping. Even though some combinations of transformations might never appear (e.g. an upright face with a horizontal nose), current equivariant architectures consider the set of all possible transformations in a transformation group when learning feature representations. Contrarily, the human visual system is able to attend to the set of relevant transformations occurring in the environment and utilizes this information to assist and improve object recognition. Based on this observation, we modify conventional equivariant feature mappings such that they are able to attend to the set of co-occurring transformations in data and generalize this notion to act on groups consisting of multiple symmetries. We show that our proposed co-attentive equivariant neural networks consistently outperform conventional rotation equivariant and rotation & reflection equivariant neural networks on rotated MNIST and CIFAR-10.

59.Modality To Modality Translation: An Adversarial Representation Learning and Graph Fusion Network for Multimodal Fusion ⬇️

Learning joint embedding space for various modalities is of vital importance for multimodal fusion. Mainstream modality fusion approaches fail to achieve this goal, leaving a modality gap which heavily affects cross-modal fusion. In this paper, we propose a novel adversarial encoder-decoder-classifier framework to learn a modality-invariant embedding space. Since the distributions of various modalities vary in nature, to reduce the modality gap, we translate the distributions of source modalities into that of target modality via their respective encoders using adversarial training. Furthermore, we exert additional constraints on embedding space by introducing reconstruction and classification losses. Then we fuse the encoded representations using hierarchical graph neural network which explicitly explores unimodal, bimodal and trimodal interactions in multi-stage. Our method achieves state-of-the-art performances on multiple datasets. Visualization of the learned embeddings suggests that the joint embedding space learned by our method is discriminative.

60.Efficient Hardware Implementation of Incremental Learning and Inference on Chip ⬇️

In this paper, we tackle the problem of incrementally learning a classifier, one example at a time, directly on chip. To this end, we propose an efficient hardware implementation of a recently introduced incremental learning procedure that achieves state-of-the-art performance by combining transfer learning with majority votes and quantization techniques. The proposed design is able to accommodate for both new examples and new classes directly on the chip. We detail the hardware implementation of the method (implemented on FPGA target) and show it requires limited resources while providing a significant acceleration compared to using a CPU.

61.Multiple Face Analyses through Adversarial Learning ⬇️

This inherent relations among multiple face analysis tasks, such as landmark detection, head pose estimation, gender recognition and face attribute estimation are crucial to boost the performance of each task, but have not been thoroughly explored since typically these multiple face analysis tasks are handled as separate tasks. In this paper, we propose a novel deep multi-task adversarial learning method to localize facial landmark, estimate head pose and recognize gender jointly or estimate multiple face attributes simultaneously through exploring their dependencies from both image representation-level and label-level. Specifically, the proposed method consists of a deep recognition network R and a discriminator D. The deep recognition network is used to learn the shared middle-level image representation and conducts multiple face analysis tasks simultaneously. Through multi-task learning mechanism, the recognition network explores the dependencies among multiple face analysis tasks, such as facial landmark localization, head pose estimation, gender recognition and face attribute estimation from image representation-level. The discriminator is introduced to enforce the distribution of the multiple face analysis tasks to converge to that inherent in the ground-truth labels. During training, the recognizer tries to confuse the discriminator, while the discriminator competes with the recognizer through distinguishing the predicted label combination from the ground-truth one. Though adversarial learning, we explore the dependencies among multiple face analysis tasks from label-level. Experimental results on four benchmark databases, i.e., the AFLW database, the Multi-PIE database, the CelebA database and the LFWA database, demonstrate the effectiveness of the proposed method for multiple face analyses.

62.Exploiting Human Social Cognition for the Detection of Fake and Fraudulent Faces via Memory Networks ⬇️

Advances in computer vision have brought us to the point where we have the ability to synthesise realistic fake content. Such approaches are seen as a source of disinformation and mistrust, and pose serious concerns to governments around the world. Convolutional Neural Networks (CNNs) demonstrate encouraging results when detecting fake images that arise from the specific type of manipulation they are trained on. However, this success has not transitioned to unseen manipulation types, resulting in a significant gap in the line-of-defense. We propose a Hierarchical Memory Network (HMN) architecture, which is able to successfully detect faked faces by utilising knowledge stored in neural memories as well as visual cues to reason about the perceived face and anticipate its future semantic embeddings. This renders a generalisable face tampering detection framework. Experimental results demonstrate the proposed approach achieves superior performance for fake and fraudulent face detection compared to the state-of-the-art.

63.Tigris: Architecture and Algorithms for 3D Perception in Point Clouds ⬇️

Machine perception applications are increasingly moving toward manipulating and processing 3D point cloud. This paper focuses on point cloud registration, a key primitive of 3D data processing widely used in high-level tasks such as odometry, simultaneous localization and mapping, and 3D reconstruction. As these applications are routinely deployed in energy-constrained environments, real-time and energy-efficient point cloud registration is critical.
We present Tigris, an algorithm-architecture co-designed system specialized for point cloud registration. Through an extensive exploration of the registration pipeline design space, we find that, while different design points make vastly different trade-offs between accuracy and performance, KD-tree search is a common performance bottleneck, and thus is an ideal candidate for architectural specialization. While KD-tree search is inherently sequential, we propose an acceleration-amenable data structure and search algorithm that exposes different forms of parallelism of KD-tree search in the context of point cloud registration. The co-designed accelerator systematically exploits the parallelism while incorporating a set of architectural techniques that further improve the accelerator efficiency. Overall, Tigris achieves 77.2$\times$ speedup and 7.4$\times$ power reduction in KD-tree search over an RTX 2080 Ti GPU, which translates to a 41.7% registration performance improvements and 3.0$\times$ power reduction.

64.LNDb: A Lung Nodule Database on Computed Tomography ⬇️

Lung cancer is the deadliest type of cancer worldwide and late detection is the major factor for the low survival rate of patients. Low dose computed tomography has been suggested as a potential screening tool but manual screening is costly, time-consuming and prone to variability. This has fueled the development of automatic methods for the detection, segmentation and characterisation of pulmonary nodules but its application to clinical routine is challenging. In this study, a new database for the development and testing of pulmonary nodule computer-aided strategies is presented which intends to complement current databases by giving additional focus to radiologist variability and local clinical reality. State-of-the-art nodule detection, segmentation and characterization methods are tested and compared to manual annotations as well as collaborative strategies combining multiple radiologists and radiologists and computer-aided systems. It is shown that state-of-the-art methodologies can determine a patient's follow-up recommendation as accurately as a radiologist, though the nodule detection method used shows decreased performance in this database.

65.Multi-Resolution 3D CNN for MRI Brain Tumor Segmentation and Survival Prediction ⬇️

In this study, an automated three dimensional (3D) deep segmentation approach for detecting gliomas in 3D pre-operative MRI scans is proposed. Then, a classi-fication algorithm based on random forests, for survival prediction is presented. The objective is to segment the glioma area and produce segmentation labels for its different sub-regions, i.e. necrotic and the non-enhancing tumor core, the peri-tumoral edema, and enhancing tumor. The proposed deep architecture for the segmentation task encompasses two parallel streamlines with two different reso-lutions. One deep convolutional neural network is to learn local features of the input data while the other one is set to have a global observation on whole image. Deemed to be complementary, the outputs of each stream are then merged to pro-vide an ensemble complete learning of the input image. The proposed network takes the whole image as input instead of patch-based approaches in order to con-sider the semantic features throughout the whole volume. The algorithm is trained on BraTS 2019 which included 335 training cases, and validated on 127 unseen cases from the validation dataset using a blind testing approach. The proposed method was also evaluated on the BraTS 2019 challenge test dataset of 166 cases. The results show that the proposed methods provide promising segmentations as well as survival prediction. The mean Dice overlap measures of automatic brain tumor segmentation for validation set were 0.84, 0.74 and 0.71 for the whole tu-mor, core and enhancing tumor, respectively. The corresponding results for the challenge test dataset were 0.82, 0.72, and 0.70, respectively. The overall accura-cy of the proposed model for the survival prediction task is %52 for the valida-tion and %49 for the test dataset.

66.Live Face De-Identification in Video ⬇️

We propose a method for face de-identification that enables fully automatic video modification at high frame rates. The goal is to maximally decorrelate the identity, while having the perception (pose, illumination and expression) fixed. We achieve this by a novel feed-forward encoder-decoder network architecture that is conditioned on the high-level representation of a person's facial image. The network is global, in the sense that it does not need to be retrained for a given video or for a given identity, and it creates natural looking image sequences with little distortion in time.

67.Exactly Sparse Gaussian Variational Inference with Application to Derivative-Free Batch Nonlinear State Estimation ⬇️

We present a Gaussian Variational Inference (GVI) technique that can be applied to large-scale nonlinear batch state estimation problems. The main contribution is to show how to fit the best Gaussian to the posterior efficiently by exploiting factorization of the joint likelihood of the state and data, as is common in practical problems. The proposed Exactly Sparse Gaussian Variational Inference (ESGVI) technique stores the inverse covariance matrix, which is typically very sparse (e.g., block-tridiagonal for classic state estimation). We show that the only blocks of the (dense) covariance matrix that are required during the calculations correspond to the non-zero blocks of the inverse covariance matrix, and further show how to calculate these blocks efficiently in the general GVI problem. ESGVI operates iteratively, and while we can use analytical derivatives at each iteration, Gaussian cubature can be substituted, thereby producing an efficient derivative-free batch formulation. ESGVI simplifies to precisely the Rauch-Tung-Striebel (RTS) smoother in the batch linear estimation case, but goes beyond the 'extended' RTS smoother in the nonlinear case since it finds the best-fit Gaussian, not the Maximum A Posteriori (MAP) point solution. We demonstrate the technique on controlled simulation problems and a batch nonlinear Simultaneous Localization and Mapping (SLAM) problem with an experimental dataset.

68.Unsupervised AER Object Recognition Based on Multiscale Spatio-Temporal Features and Spiking Neurons ⬇️

This paper proposes an unsupervised address event representation (AER) object recognition approach. The proposed approach consists of a novel multiscale spatio-temporal feature (MuST) representation of input AER events and a spiking neural network (SNN) using spike-timing-dependent plasticity (STDP) for object recognition with MuST. MuST extracts the features contained in both the spatial and temporal information of AER event flow, and meanwhile forms an informative and compact feature spike representation. We show not only how MuST exploits spikes to convey information more effectively, but also how it benefits the recognition using SNN. The recognition process is performed in an unsupervised manner, which does not need to specify the desired status of every single neuron of SNN, and thus can be flexibly applied in real-world recognition tasks. The experiments are performed on five AER datasets including a new one named GESTURE-DVS. Extensive experimental results show the effectiveness and advantages of this proposed approach.

69.Projection-to-Projection Translation for Hybrid X-ray and Magnetic Resonance Imaging ⬇️

Hybrid X-ray and magnetic resonance (MR) imaging promises large potential in interventional medical imaging applications due to the broad variety of contrast of MRI combined with fast imaging of X-ray-based modalities. To fully utilize the potential of the vast amount of existing image enhancement techniques, the corresponding information from both modalities must be present in the same domain. For image-guided interventional procedures, X-ray fluoroscopy has proven to be the modality of choice. Synthesizing one modality from another in this case is an ill-posed problem due to ambiguous signal and overlapping structures in projective geometry. To take on these challenges, we present a learning-based solution to MR to X-ray projection-to-projection translation. We propose an image generator network that focuses on high representation capacity in higher resolution layers to allow for accurate synthesis of fine details in the projection images. Additionally, a weighting scheme in the loss computation that favors high-frequency structures is proposed to focus on the important details and contours in projection imaging. The proposed extensions prove valuable in generating X-ray projection images with natural appearance. Our approach achieves a deviation from the ground truth of only $6$% and structural similarity measure of $0.913,\pm,0.005$. In particular the high frequency weighting assists in generating projection images with sharp appearance and reduces erroneously synthesized fine details.

70.Estimation of Orientation and Camera Parameters from Cryo-Electron Microscopy Images with Variational Autoencoders and Generative Adversarial Networks ⬇️

Cryo-electron microscopy (cryo-EM) is capable of producing reconstructed 3D images of biomolecules at near-atomic resolution. As such, it represents one of the most promising imaging techniques in structural biology. However, raw cryo-EM images are only highly corrupted - noisy and band-pass filtered - 2D projections of the target 3D biomolecules. Reconstructing the 3D molecular shape starts with the removal of image outliers, the estimation of the orientation of the biomolecule that has produced the given 2D image, and the estimation of camera parameters to correct for intensity defects. Current techniques performing these tasks are often computationally expensive, while the dataset sizes keep growing. There is a need for next-generation algorithms that preserve accuracy while improving speed and scalability. In this paper, we combine variational autoencoders (VAEs) and generative adversarial networks (GANs) to learn a low-dimensional latent representation of cryo-EM images. We perform an exploratory analysis of the obtained latent space, that is shown to have a structure of "orbits", in the sense of Lie group theory, consistent with the acquisition procedure of cryo-EM images. This analysis leads us to design an estimation method for orientation and camera parameters of single-particle cryo-EM images, together with an outliers detection procedure. As such, it opens the door to geometric approaches for unsupervised estimations of orientations and camera parameters, making possible fast cryo-EM biomolecule reconstruction.

71.Adaptive Routing Between Capsules ⬇️

Capsule network is the most recent exciting advancement in the deep learning field and represents positional information by stacking features into vectors. The dynamic routing algorithm is used in the capsule network, however, there are some disadvantages such as the inability to stack multiple layers and a large amount of computation. In this paper, we propose an adaptive routing algorithm that can solve the problems mentioned above. First, the low-layer capsules adaptively adjust their direction and length in the routing algorithm and removing the influence of the coupling coefficient on the gradient propagation, so that the network can work when stacked in multiple layers. Then, the iterative process of routing is simplified to reduce the amount of computation and we introduce the gradient coefficient $\lambda$. Further, we tested the performance of our proposed adaptive routing algorithm on CIFAR10, Fashion-MNIST, SVHN and MNIST, while achieving better results than the dynamic routing algorithm.

72.Three-dimensional Generative Adversarial Nets for Unsupervised Metal Artifact Reduction ⬇️

The reduction of metal artifacts in computed tomography (CT) images, specifically for strong artifacts generated from multiple metal objects, is a challenging issue in medical imaging research. Although there have been some studies on supervised metal artifact reduction through the learning of synthesized artifacts, it is difficult for simulated artifacts to cover the complexity of the real physical phenomena that may be observed in X-ray propagation. In this paper, we introduce metal artifact reduction methods based on an unsupervised volume-to-volume translation learned from clinical CT images. We construct three-dimensional adversarial nets with a regularized loss function designed for metal artifacts from multiple dental fillings. The results of experiments using 915 CT volumes from real patients demonstrate that the proposed framework has an outstanding capacity to reduce strong artifacts and to recover underlying missing voxels, while preserving the anatomical features of soft tissues and tooth structures from the original images.

73.HighEr-Resolution Network for Image Demosaicing and Enhancing ⬇️

Neural-networks based image restoration methods tend to use low-resolution image patches for training. Although higher-resolution image patches can provide more global information, state-of-the-art methods cannot utilize them due to their huge GPU memory usage, as well as the instable training process. However, plenty of studies have shown that global information is crucial for image restoration tasks like image demosaicing and enhancing. In this work, we propose a HighEr-Resolution Network (HERN) to fully learning global information in high-resolution image patches. To achieve this, the HERN employs two parallel paths to learn image features in two different resolutions, respectively. By combining global-aware features and multi-scale features, our HERN is able to learn global information with feasible GPU memory usage. Besides, we introduce a progressive training method to solve the instability issue and accelerate model convergence. On the task of image demosaicing and enhancing, our HERN achieves state-of-the-art performance on the AIM2019 RAW to RGB mapping challenge. The source code of our implementation is available at this https URL.

74.AddNet: Deep Neural Networks Using FPGA-Optimized Multipliers ⬇️

Low-precision arithmetic operations to accelerate deep-learning applications on field-programmable gate arrays (FPGAs) have been studied extensively, because they offer the potential to save silicon area or increase throughput. However, these benefits come at the cost of a decrease in accuracy. In this article, we demonstrate that reconfigurable constant coefficient multipliers (RCCMs) offer a better alternative for saving the silicon area than utilizing low-precision arithmetic. RCCMs multiply input values by a restricted choice of coefficients using only adders, subtractors, bit shifts, and multiplexers (MUXes), meaning that they can be heavily optimized for FPGAs. We propose a family of RCCMs tailored to FPGA logic elements to ensure their efficient utilization. To minimize information loss from quantization, we then develop novel training techniques that map the possible coefficient representations of the RCCMs to neural network weight parameter distributions. This enables the usage of the RCCMs in hardware, while maintaining high accuracy. We demonstrate the benefits of these techniques using AlexNet, ResNet-18, and ResNet-50 networks. The resulting implementations achieve up to 50% resource savings over traditional 8-bit quantized networks, translating to significant speedups and power savings. Our RCCM with the lowest resource requirements exceeds 6-bit fixed point accuracy, while all other implementations with RCCMs achieve at least similar accuracy to an 8-bit uniformly quantized design, while achieving significant resource savings.

75.Deep Detector Health Management under Adversarial Campaigns ⬇️

Machine learning models are vulnerable to adversarial inputs that induce seemingly unjustifiable errors. As automated classifiers are increasingly used in industrial control systems and machinery, these adversarial errors could grow to be a serious problem. Despite numerous studies over the past few years, the field of adversarial ML is still considered alchemy, with no practical unbroken defenses demonstrated to date, leaving PHM practitioners with few meaningful ways of addressing the problem. We introduce turbidity detection as a practical superset of the adversarial input detection problem, coping with adversarial campaigns rather than statistically invisible one-offs. This perspective is coupled with ROC-theoretic design guidance that prescribes an inexpensive domain adaptation layer at the output of a deep learning model during an attack campaign. The result aims to approximate the Bayes optimal mitigation that ameliorates the detection model's degraded health. A proactively reactive type of prognostics is achieved via Monte Carlo simulation of various adversarial campaign scenarios, by sampling from the model's own turbidity distribution to quickly deploy the correct mitigation during a real-world campaign.

76.Online Learned Continual Compression with Stacked Quantization Module ⬇️

We introduce and study the problem of Online Continual Compression, where one attempts to learn to compress and store a representative dataset from a non i.i.d data stream, while only observing each sample once. This problem is highly relevant for downstream online continual learning tasks, as well as standard learning methods under resource constrained data collection. To address this we propose a new architecture which Stacks Quantization Modules (SQM), consisting of a series of discrete autoencoders, each equipped with their own memory. Every added module is trained to reconstruct the latent space of the previous module using fewer bits, allowing the learned representation to become more compact as training progresses. This modularity has several advantages: 1) moderate compressions are quickly available early in training, which is crucial for remembering the early tasks, 2) as more data needs to be stored, earlier data becomes more compressed, freeing memory, 3) unlike previous methods, our approach does not require pretraining, even on challenging datasets. We show several potential applications of this method. We first replace the episodic memory used in Experience Replay with SQM, leading to significant gains on standard continual learning benchmarks using a fixed memory budget. We then apply our method to online compression of larger images like those from Imagenet, and show that it is also effective with other modalities, such as LiDAR data.

77.Convolutional Neural Network and decision support in medical imaging: case study of the recognition of blood cell subtypes ⬇️

Identifying and characterizing the patient's blood samples is indispensable in diagnostics of malignance suspicious. A painstaking and sometimes subjective task is used in laboratories to manually classify white blood cells. Neural mathematical methods as deep learnings can be very useful in the automated recognition of blood cells. This study uses a particular type of deep learning i.e., convolutional neural networks (CNNs or ConvNets) for image recognition of the four (4) blood cell types (neutrophil, eosinophil, lymphocyte and monocyte) and to enable it to tag them employing a dataset of blood cells with labels for the corresponding cell types. The elements of the database are the input of our CNN and they allowed us to create learning models for the image recognition/classification of the blood cells. We evaluated the recognition performance and outputs learned by the networks in order to implement a neural image recognition model capable of distinguishing polynuclear cells (neutrophil and eosinophil) from those of mononuclear cells (lymphocyte and monocyte). The validation accuracy is 97.77%.

78.CD2 : Combined Distances of Contrast Distributions for the Assessment of Perceptual Quality of Image Processing ⬇️

The quality of visual input is very important for both human and machine perception. Consequently many processing techniques exist that deal with different distortions. Usually image processing is applied freely and lacks redundancy regarding safety. We propose a novel image comparison method called the Combined Distances of Contrast Distributions (CD2) to protect against errors that arise during processing. Based on the distribution of image contrasts a new reduced-reference image quality assessment (IQA) method is introduced. By combining various distance functions excellent performance on IQA benchmarks is achieved with only a small data and computation overhead.

79.WITCHcraft: Efficient PGD attacks with random step size ⬇️

State-of-the-art adversarial attacks on neural networks use expensive iterative methods and numerous random restarts from different initial points. Iterative FGSM-based methods without restarts trade off performance for computational efficiency because they do not adequately explore the image space and are highly sensitive to the choice of step size. We propose a variant of Projected Gradient Descent (PGD) that uses a random step size to improve performance without resorting to expensive random restarts. Our method, Wide Iterative Stochastic crafting (WITCHcraft), achieves results superior to the classical PGD attack on the CIFAR-10 and MNIST data sets but without additional computational cost. This simple modification of PGD makes crafting attacks more economical, which is important in situations like adversarial training where attacks need to be crafted in real time.

80.Learning Permutation Invariant Representations using Memory Networks ⬇️

Many real world tasks such as 3D object detection and high-resolution image classification involve learning from a set of instances. In these cases, only a group of instances, a set, collectively contains meaningful information and therefore only the sets have labels, and not individual data instances. In this work, we present a permutation invariant neural network called a \textbf{Memory-based Exchangeable Model (MEM)} for learning set functions. The model consists of memory units that embed an input sequence to high-level features (memories) enabling the model to learn inter-dependencies among instances of the set in the form of attention vectors. To demonstrate its learning ability, we evaluated our model on test datasets created using MNIST, point cloud classification, and population estimation. We also tested the model for classifying histopathology whole slide images to discriminate between two subtypes of Lung cancer---Lung Adenocarcinoma, and Lung Squamous Cell Carcinoma. We systematically extracted patches from lung cancer images from The Cancer Genome Atlas~(TCGA) dataset, the largest public repository of histopathology images. The proposed method achieved a competitive classification accuracy of 84.84%. The results on other datasets are promising and demonstrate the efficacy of our model.

81.Unsupervised Domain Adaptation via Structured Prediction Based Selective Pseudo-Labeling ⬇️

Unsupervised domain adaptation aims to address the problem of classifying unlabeled samples from the target domain whilst labeled samples are only available from the source domain and the data distributions are different in these two domains. As a result, classifiers trained from labeled samples in the source domain suffer from significant performance drop when directly applied to the samples from the target domain. To address this issue, different approaches have been proposed to learn domain-invariant features or domain-specific classifiers. In either case, the lack of labeled samples in the target domain can be an issue which is usually overcome by pseudo-labeling. Inaccurate pseudo-labeling, however, could result in catastrophic error accumulation during learning. In this paper, we propose a novel selective pseudo-labeling strategy based on structured prediction. The idea of structured prediction is inspired by the fact that samples in the target domain are well clustered within the deep feature space so that unsupervised clustering analysis can be used to facilitate accurate pseudo-labeling. Experimental results on four datasets (i.e. Office-Caltech, Office31, ImageCLEF-DA and Office-Home) validate our approach outperforms contemporary state-of-the-art methods.

82.Implicit Regularization of Normalization Methods ⬇️

Normalization methods such as batch normalization are commonly used in overparametrized models like neural networks. Here, we study the weight normalization (WN) method (Salimans & Kingma, 2016) and a variant called reparametrized projected gradient descent (rPGD) for overparametrized least squares regression and some more general loss functions. WN and rPGD reparametrize the weights with a scale $g$ and a unit vector such that the objective function becomes \emph{non-convex}. We show that this non-convex formulation has beneficial regularization effects compared to gradient descent on the original objective. We show that these methods adaptively regularize the weights and \emph{converge with exponential rate} to the minimum $\ell_2$ norm solution (or close to it) even for initializations \emph{far from zero}. This is different from the behavior of gradient descent, which only converges to the min norm solution when started at zero, and is more sensitive to initialization. Some of our proof techniques are different from many related works; for instance we find explicit invariants along the gradient flow paths. We verify our results experimentally and suggest that there may be a similar phenomenon for nonlinear problems such as matrix sensing.

83.ISP4ML: Understanding the Role of Image Signal Processing in Efficient Deep Learning Vision Systems ⬇️

Convolutional neural networks (CNNs) are now predominant components in a variety of computer vision (CV) systems. These systems typically include an image signal processor (ISP), even though the ISP is traditionally designed to produce images that look appealing to humans. In CV systems, it is not clear what the role of the ISP is, or if it is even required at all for accurate prediction. In this work, we investigate the efficacy of the ISP in CNN classification tasks, and outline the system-level trade-offs between prediction accuracy and computational cost. To do so, we build software models of a configurable ISP and an imaging sensor in order to train CNNs on ImageNet with a range of different ISP settings and functionality. Results on ImageNet show that an ISP improves accuracy by 4.6%-12.2% on MobileNet architectures of different widths. Results using ResNets demonstrate that these trends also generalize to deeper networks. An ablation study of the various processing stages in a typical ISP reveals that the tone mapper is the most significant stage when operating on high dynamic range (HDR) images, by providing 5.8% average accuracy improvement alone. Overall, the ISP benefits system efficiency because the memory and computational costs of the ISP is minimal compared to the cost of using a larger CNN to achieve the same accuracy.

84.Deep Learning Captures More Accurate Diffusion Fiber Orientations Distributions than Constrained Spherical Deconvolution ⬇️

Confocal histology provides an opportunity to establish intra-voxel fiber orientation distributions that can be used to quantitatively assess the biological relevance of diffusion weighted MRI models, e.g., constrained spherical deconvolution (CSD). Here, we apply deep learning to investigate the potential of single shell diffusion weighted MRI to explain histologically observed fiber orientation distributions (FOD) and compare the derived deep learning model with a leading CSD approach. This study (1) demonstrates that there exists additional information in the diffusion signal that is not currently exploited by CSD, and (2) provides an illustrative data-driven model that makes use of this information.

85.A Deep Learning Approach for Robust Corridor Following ⬇️

For an autonomous corridor following task where the environment is continuously changing, several forms of environmental noise prevent an automated feature extraction procedure from performing reliably. Moreover, in cases where pre-defined features are absent from the captured data, a well defined control signal for performing the servoing task fails to get produced. In order to overcome these drawbacks, we present in this work, using a convolutional neural network (CNN) to directly estimate the required control signal from an image, encompassing feature extraction and control law computation into one single end-to-end framework. In particular, we study the task of autonomous corridor following using a CNN and present clear advantages in cases where a traditional method used for performing the same task fails to give a reliable outcome. We evaluate the performance of our method on this task on a Wheelchair Platform developed at our institute for this purpose.

86.Frequency Separation for Real-World Super-Resolution ⬇️

Most of the recent literature on image super-resolution (SR) assumes the availability of training data in the form of paired low resolution (LR) and high resolution (HR) images or the knowledge of the downgrading operator (usually bicubic downscaling). While the proposed methods perform well on standard benchmarks, they often fail to produce convincing results in real-world settings. This is because real-world images can be subject to corruptions such as sensor noise, which are severely altered by bicubic downscaling. Therefore, the models never see a real-world image during training, which limits their generalization capabilities. Moreover, it is cumbersome to collect paired LR and HR images in the same source domain.
To address this problem, we propose DSGAN to introduce natural image characteristics in bicubically downscaled images. It can be trained in an unsupervised fashion on HR images, thereby generating LR images with the same characteristics as the original images. We then use the generated data to train a SR model, which greatly improves its performance on real-world images. Furthermore, we propose to separate the low and high image frequencies and treat them differently during training. Since the low frequencies are preserved by downsampling operations, we only require adversarial training to modify the high frequencies. This idea is applied to our DSGAN model as well as the SR model. We demonstrate the effectiveness of our method in several experiments through quantitative and qualitative analysis. Our solution is the winner of the AIM Challenge on Real World SR at ICCV 2019.

87.Neural Forest Learning ⬇️

We propose Neural Forest Learning (NFL), a novel deep learning based random-forest-like method. In contrast to previous forest methods, NFL enjoys the benefits of end-to-end, data-driven representation learning, as well as pervasive support from deep learning software and hardware platforms, hence achieving faster inference speed and higher accuracy than previous forest methods. Furthermore, NFL learns non-linear feature representations in CNNs more efficiently than previous higher-order pooling methods, producing good results with negligible increase in parameters, floating point operations (FLOPs) and real running time. We achieve superior performance on 7 machine learning datasets when compared to random forests and GBDTs. On the fine-grained benchmarks CUB-200-2011, FGVC-aircraft and Stanford Cars, we achieve over 5.7%, 6.9% and 7.8% gains for VGG-16, respectively. Moreover, NFL can converge in much fewer epochs, further accelerating network training. On the large-scale ImageNet ILSVRC-12 validation set, integration of NFL into ResNet-18 achieves top-1/top-5 errors of 28.32%/9.77%, which outperforms ResNet-18 by 1.92%/1.15% with negligible extra cost and the improvement is consistent under various architectures.