20220708.md

ArXiv cs.CV --Fri, 8 Jul 2022

1.Finding Fallen Objects Via Asynchronous Audio-Visual Integration ⬇️

The way an object looks and sounds provide complementary reflections of its physical properties. In many settings cues from vision and audition arrive asynchronously but must be integrated, as when we hear an object dropped on the floor and then must find it. In this paper, we introduce a setting in which to study multi-modal object localization in 3D virtual environments. An object is dropped somewhere in a room. An embodied robot agent, equipped with a camera and microphone, must determine what object has been dropped -- and where -- by combining audio and visual signals with knowledge of the underlying physics. To study this problem, we have generated a large-scale dataset -- the Fallen Objects dataset -- that includes 8000 instances of 30 physical object categories in 64 rooms. The dataset uses the ThreeDWorld platform which can simulate physics-based impact sounds and complex physical interactions between objects in a photorealistic setting. As a first step toward addressing this challenge, we develop a set of embodied agent baselines, based on imitation learning, reinforcement learning, and modular planning, and perform an in-depth analysis of the challenge of this new task.

2.Bridging the Gap between Object and Image-level Representations for Open-Vocabulary Detection ⬇️

Existing open-vocabulary object detectors typically enlarge their vocabulary sizes by leveraging different forms of weak supervision. This helps generalize to novel objects at inference. Two popular forms of weak-supervision used in open-vocabulary detection (OVD) include pretrained CLIP model and image-level supervision. We note that both these modes of supervision are not optimally aligned for the detection task: CLIP is trained with image-text pairs and lacks precise localization of objects while the image-level supervision has been used with heuristics that do not accurately specify local object regions. In this work, we propose to address this problem by performing object-centric alignment of the language embeddings from the CLIP model. Furthermore, we visually ground the objects with only image-level supervision using a pseudo-labeling process that provides high-quality object proposals and helps expand the vocabulary during training. We establish a bridge between the above two object-alignment strategies via a novel weight transfer function that aggregates their complimentary strengths. In essence, the proposed model seeks to minimize the gap between object and image-centric representations in the OVD setting. On the COCO benchmark, our proposed approach achieves 40.3 AP50 on novel classes, an absolute 11.9 gain over the previous best performance.For LVIS, we surpass the state-of-the-art ViLD model by 5.0 mask AP for rare categories and 3.4 overall. Code: this https URL.

3.Red PANDA: Disambiguating Anomaly Detection by Removing Nuisance Factors ⬇️

Anomaly detection methods strive to discover patterns that differ from the norm in a semantic way. This goal is ambiguous as a data point differing from the norm by an attribute e.g., age, race or gender, may be considered anomalous by some operators while others may consider this attribute irrelevant. Breaking from previous research, we present a new anomaly detection method that allows operators to exclude an attribute from being considered as relevant for anomaly detection. Our approach then learns representations which do not contain information over the nuisance attributes. Anomaly scoring is performed using a density-based approach. Importantly, our approach does not require specifying the attributes that are relevant for detecting anomalies, which is typically impossible in anomaly detection, but only attributes to ignore. An empirical investigation is presented verifying the effectiveness of our approach.

4.LASSIE: Learning Articulated Shapes from Sparse Image Ensemble via 3D Part Discovery ⬇️

Creating high-quality articulated 3D models of animals is challenging either via manual creation or using 3D scanning tools. Therefore, techniques to reconstruct articulated 3D objects from 2D images are crucial and highly useful. In this work, we propose a practical problem setting to estimate 3D pose and shape of animals given only a few (10-30) in-the-wild images of a particular animal species (say, horse). Contrary to existing works that rely on pre-defined template shapes, we do not assume any form of 2D or 3D ground-truth annotations, nor do we leverage any multi-view or temporal information. Moreover, each input image ensemble can contain animal instances with varying poses, backgrounds, illuminations, and textures. Our key insight is that 3D parts have much simpler shape compared to the overall animal and that they are robust w.r.t. animal pose articulations. Following these insights, we propose LASSIE, a novel optimization framework which discovers 3D parts in a self-supervised manner with minimal user intervention. A key driving force behind LASSIE is the enforcing of 2D-3D part consistency using self-supervisory deep features. Experiments on Pascal-Part and self-collected in-the-wild animal datasets demonstrate considerably better 3D reconstructions as well as both 2D and 3D part discovery compared to prior arts. Project page: this http URL

5.Semi-supervised Object Detection via Virtual Category Learning ⬇️

Due to the costliness of labelled data in real-world applications, semi-supervised object detectors, underpinned by pseudo labelling, are appealing. However, handling confusing samples is nontrivial: discarding valuable confusing samples would compromise the model generalisation while using them for training would exacerbate the confirmation bias issue caused by inevitable mislabelling. To solve this problem, this paper proposes to use confusing samples proactively without label correction. Specifically, a virtual category (VC) is assigned to each confusing sample such that they can safely contribute to the model optimisation even without a concrete label. It is attributed to specifying the embedding distance between the training sample and the virtual category as the lower bound of the inter-class distance. Moreover, we also modify the localisation loss to allow high-quality boundaries for location regression. Extensive experiments demonstrate that the proposed VC learning significantly surpasses the state-of-the-art, especially with small amounts of available labels.

6.Domain Knowledge Driven 3D Dose Prediction Using Moment-Based Loss Function ⬇️

Dose volume histogram (DVH) metrics are widely accepted evaluation criteria in the clinic. However, incorporating these metrics into deep learning dose prediction models is challenging due to their non-convexity and non-differentiability. We propose a novel moment-based loss function for predicting 3D dose distribution for the challenging conventional lung intensity modulated radiation therapy (IMRT) plans. The moment-based loss function is convex and differentiable and can easily incorporate DVH metrics in any deep learning framework without computational overhead. The moments can also be customized to reflect the clinical priorities in 3D dose prediction. For instance, using high-order moments allows better prediction in high-dose areas for serial structures. We used a large dataset of 360 (240 for training, 50 for validation and 70 for testing) conventional lung patients with 2Gy $\times$ 30 fractions to train the deep learning (DL) model using clinically treated plans at our institution. We trained a UNet like CNN architecture using computed tomography (CT), planning target volume (PTV) and organ-at-risk contours (OAR) as input to infer corresponding voxel-wise 3D dose distribution. We evaluated three different loss functions: (1) The popular Mean Absolute Error (MAE) Loss, (2) the recently developed MAE + DVH Loss, and (3) the proposed MAE + Moments Loss. The quality of the predictions was compared using different DVH metrics as well as dose-score and DVH-score, recently introduced by the AAPM knowledge-based planning grand challenge. Model with (MAE + Moment) loss function outperformed the model with MAE loss by significantly improving the DVH-score (11%, p$<$0.01) while having similar computational cost. It also outperformed the model trained with (MAE+DVH) by significantly improving the computational cost (48%) and the DVH-score (8%, p$<$0.01).

7.VecGAN: Image-to-Image Translation with Interpretable Latent Directions ⬇️

We propose VecGAN, an image-to-image translation framework for facial attribute editing with interpretable latent directions. Facial attribute editing task faces the challenges of precise attribute editing with controllable strength and preservation of the other attributes of an image. For this goal, we design the attribute editing by latent space factorization and for each attribute, we learn a linear direction that is orthogonal to the others. The other component is the controllable strength of the change, a scalar value. In our framework, this scalar can be either sampled or encoded from a reference image by projection. Our work is inspired by the latent space factorization works of fixed pretrained GANs. However, while those models cannot be trained end-to-end and struggle to edit encoded images precisely, VecGAN is end-to-end trained for image translation task and successful at editing an attribute while preserving the others. Our extensive experiments show that VecGAN achieves significant improvements over state-of-the-arts for both local and global edits.

8.Robust Watermarking for Video Forgery Detection with Improved Imperceptibility and Robustness ⬇️

Videos are prone to tampering attacks that alter the meaning and deceive the audience. Previous video forgery detection schemes find tiny clues to locate the tampered areas. However, attackers can successfully evade supervision by destroying such clues using video compression or blurring. This paper proposes a video watermarking network for tampering localization. We jointly train a 3D-UNet-based watermark embedding network and a decoder that predicts the tampering mask. The perturbation made by watermark embedding is close to imperceptible. Considering that there is no off-the-shelf differentiable video codec simulator, we propose to mimic video compression by ensembling simulation results of other typical attacks, e.g., JPEG compression and blurring, as an approximation. Experimental results demonstrate that our method generates watermarked videos with good imperceptibility and robustly and accurately locates tampered areas within the attacked version.

9.Diagnosing and Remedying Shot Sensitivity with Cosine Few-Shot Learners ⬇️

Few-shot recognition involves training an image classifier to distinguish novel concepts at test time using few examples (shot). Existing approaches generally assume that the shot number at test time is known in advance. This is not realistic, and the performance of a popular and foundational method has been shown to suffer when train and test shots do not match. We conduct a systematic empirical study of this phenomenon. In line with prior work, we find that shot sensitivity is broadly present across metric-based few-shot learners, but in contrast to prior work, larger neural architectures provide a degree of built-in robustness to varying test shot. More importantly, a simple, previously known but greatly overlooked class of approaches based on cosine distance consistently and greatly improves robustness to shot variation, by removing sensitivity to sample noise. We derive cosine alternatives to popular and recent few-shot classifiers, broadening their applicability to realistic settings. These cosine models consistently improve shot-robustness, outperform prior shot-robust state of the art, and provide competitive accuracy on a range of benchmarks and architectures, including notable gains in the very-low-shot regime.

10.Joint Super-Resolution and Inverse Tone-Mapping: A Feature Decomposition Aggregation Network and A New Benchmark ⬇️

Joint Super-Resolution and Inverse Tone-Mapping (joint SR-ITM) aims to increase the resolution and dynamic range of low-resolution and standard dynamic range images.Recent methods mainly resort to image decomposition techniques with the multi-branch network architecture.However, the rigid decomposition employed by these methods largely restricts their power on diverse this http URL exploit its potential power, in this paper, we generalize the decomposition mechanism from the image domain to the broader feature domain. To this end, we propose a lightweight Feature Decomposition Aggregation Network (FDAN). In particular, we design a Feature Decomposition Block (FDB), which can achieve learnable separation of feature details and this http URL cascading FDBs, we can build up a Hierarchical Feature Decomposition Group for powerful multi-level feature decomposition.Moreover, we collect a new benchmark dataset for joint SR-ITM, \ie, SRITM-4K, which is large-scale and provides versatile scenarios for sufficient model training and evaluation.Experimental results on two benchmark datasets demonstrate that our FDAN is efficient and outperforms previous methods on joint SR-ITM.Our code and dataset will be publicly released.

11.A simple normalization technique using window statistics to improve the out-of-distribution generalization in medical images ⬇️

Since data scarcity and data heterogeneity are prevailing for medical images, well-trained Convolutional Neural Networks (CNNs) using previous normalization methods may perform poorly when deployed to a new site. However, a reliable model for real-world applications should be able to generalize well both on in-distribution (IND) and out-of-distribution (OOD) data (e.g., the new site data). In this study, we present a novel normalization technique called window normalization (WIN), which is a simple yet effective alternative to existing normalization methods. Specifically, WIN perturbs the normalizing statistics with the local statistics computed on a window of features. This feature-level augmentation technique regularizes the models well and improves their OOD generalization significantly. Taking its advantage, we propose a novel self-distillation method called WIN-WIN to further improve the OOD generalization in classification. WIN-WIN is easily implemented with twice forward passes and a consistency constraint, which can be a simple extension for existing methods. Extensive experimental results on various tasks (such as glaucoma detection, breast cancer detection, chromosome classification, optic disc and cup segmentation, etc.) and datasets (26 datasets) demonstrate the generality and effectiveness of our methods. The code is available at this https URL.

12.Automated Classification of General Movements in Infants Using a Two-stream Spatiotemporal Fusion Network ⬇️

The assessment of general movements (GMs) in infants is a useful tool in the early diagnosis of neurodevelopmental disorders. However, its evaluation in clinical practice relies on visual inspection by experts, and an automated solution is eagerly awaited. Recently, video-based GMs classification has attracted attention, but this approach would be strongly affected by irrelevant information, such as background clutter in the video. Furthermore, for reliability, it is necessary to properly extract the spatiotemporal features of infants during GMs. In this study, we propose an automated GMs classification method, which consists of preprocessing networks that remove unnecessary background information from GMs videos and adjust the infant's body position, and a subsequent motion classification network based on a two-stream structure. The proposed method can efficiently extract the essential spatiotemporal features for GMs classification while preventing overfitting to irrelevant information for different recording environments. We validated the proposed method using videos obtained from 100 infants. The experimental results demonstrate that the proposed method outperforms several baseline models and the existing methods.

13.Monkeypox Skin Lesion Detection Using Deep Learning Models: A Feasibility Study ⬇️

The recent monkeypox outbreak has become a public health concern due to its rapid spread in more than 40 countries outside Africa. Clinical diagnosis of monkeypox in an early stage is challenging due to its similarity with chickenpox and measles. In cases where the confirmatory Polymerase Chain Reaction (PCR) tests are not readily available, computer-assisted detection of monkeypox lesions could be beneficial for surveillance and rapid identification of suspected cases. Deep learning methods have been found effective in the automated detection of skin lesions, provided that sufficient training examples are available. However, as of now, such datasets are not available for the monkeypox disease. In the current study, we first develop the ``Monkeypox Skin Lesion Dataset (MSLD)" consisting skin lesion images of monkeypox, chickenpox, and measles. The images are mainly collected from websites, news portals, and publicly accessible case reports. Data augmentation is used to increase the sample size, and a 3-fold cross-validation experiment is set up. In the next step, several pre-trained deep learning models, namely, VGG-16, ResNet50, and InceptionV3 are employed to classify monkeypox and other diseases. An ensemble of the three models is also developed. ResNet50 achieves the best overall accuracy of $82.96(\pm4.57%)$, while VGG16 and the ensemble system achieved accuracies of $81.48(\pm6.87%)$ and $79.26(\pm1.05%)$, respectively. A prototype web-application is also developed as an online monkeypox screening tool. While the initial results on this limited dataset are promising, a larger demographically diverse dataset is required to further enhance the generalizability of these models.

14.Softmax-free Linear Transformers ⬇️

Vision transformers (ViTs) have pushed the state-of-the-art for various visual recognition tasks by patch-wise image tokenization followed by stacked self-attention operations. Employing self-attention modules results in a quadratic complexity in both computation and memory usage. Various attempts on approximating the self-attention computation with linear complexity have thus been made in Natural Language Processing. However, an in-depth analysis in this work reveals that they are either theoretically flawed or empirically ineffective for visual recognition. We identify that their limitations are rooted in retaining the softmax self-attention during approximations. Specifically, conventional self-attention is computed by normalizing the scaled dot-product between token feature vectors. Preserving the softmax operation challenges any subsequent linearization efforts. Under this insight, a SOftmax-Free Transformer (abbreviated as SOFT) is proposed for the first time. To eliminate the softmax operator in self-attention, a Gaussian kernel function is adopted to replace the dot-product similarity. This enables a full self-attention matrix to be approximated via a low-rank matrix decomposition. The robustness of our approximation is achieved by calculating its Moore-Penrose inverse using a Newton-Raphson method. Further, an efficient symmetric normalization is introduced on the low-rank self-attention for enhancing model generalizability and transferability. Extensive experiments on ImageNet, COCO and ADE20K show that our SOFT significantly improves the computational efficiency of existing ViT variants. Crucially, with a linear complexity, much longer token sequences are permitted in SOFT, resulting in superior trade-off between accuracy and complexity.

15.Factorizing Knowledge in Neural Networks ⬇️

In this paper, we explore a novel and ambitious knowledge-transfer task, termed Knowledge Factorization~(KF). The core idea of KF lies in the modularization and assemblability of knowledge: given a pretrained network model as input, KF aims to decompose it into several factor networks, each of which handles only a dedicated task and maintains task-specific knowledge factorized from the source network. Such factor networks are task-wise disentangled and can be directly assembled, without any fine-tuning, to produce the more competent combined-task networks. In other words, the factor networks serve as Lego-brick-like building blocks, allowing us to construct customized networks in a plug-and-play manner. Specifically, each factor network comprises two modules, a common-knowledge module that is task-agnostic and shared by all factor networks, alongside with a task-specific module dedicated to the factor network itself. We introduce an information-theoretic objective, InfoMax-Bottleneck~(IMB), to carry out KF by optimizing the mutual information between the learned representations and input. Experiments across various benchmarks demonstrate that, the derived factor networks yield gratifying performances on not only the dedicated tasks but also disentanglement, while enjoying much better interpretability and modularity. Moreover, the learned common-knowledge representations give rise to impressive results on transfer learning.

16.Distilling Ensemble of Explanations for Weakly-Supervised Pre-Training of Image Segmentation Models ⬇️

While fine-tuning pre-trained networks has become a popular way to train image segmentation models, such backbone networks for image segmentation are frequently pre-trained using image classification source datasets, e.g., ImageNet. Though image classification datasets could provide the backbone networks with rich visual features and discriminative ability, they are incapable of fully pre-training the target model (i.e., backbone+segmentation modules) in an end-to-end manner. The segmentation modules are left to random initialization in the fine-tuning process due to the lack of segmentation labels in classification datasets. In our work, we propose a method that leverages Pseudo Semantic Segmentation Labels (PSSL), to enable the end-to-end pre-training for image segmentation models based on classification datasets. PSSL was inspired by the observation that the explanation results of classification models, obtained through explanation algorithms such as CAM, SmoothGrad and LIME, would be close to the pixel clusters of visual objects. Specifically, PSSL is obtained for each image by interpreting the classification results and aggregating an ensemble of explanations queried from multiple classifiers to lower the bias caused by single models. With PSSL for every image of ImageNet, the proposed method leverages a weighted segmentation learning procedure to pre-train the segmentation network en masse. Experiment results show that, with ImageNet accompanied by PSSL as the source dataset, the proposed end-to-end pre-training strategy successfully boosts the performance of various segmentation models, i.e., PSPNet-ResNet50, DeepLabV3-ResNet50, and OCRNet-HRNetW18, on a number of segmentation tasks, such as CamVid, VOC-A, VOC-C, ADE20K, and CityScapes, with significant improvements. The source code is availabel at this https URL.

17.FewSOL: A Dataset for Few-Shot Object Learning in Robotic Environments ⬇️

We introduce the Few-Shot Object Learning (FewSOL) dataset for object recognition with a few images per object. We captured 336 real-world objects with 9 RGB-D images per object from different views. Object segmentation masks, object poses and object attributes are provided. In addition, synthetic images generated using 330 3D object models are used to augment the dataset. We investigated (i) few-shot object classification and (ii) joint object segmentation and few-shot classification with the state-of-the-art methods for few-shot learning and meta-learning using our dataset. The evaluation results show that there is still a large margin to be improved for few-shot object classification in robotic environments. Our dataset can be used to study a set of few-shot object recognition problems such as classification, detection and segmentation, shape reconstruction, pose estimation, keypoint correspondences and attribute recognition. The dataset and code are available at this https URL.

18.Text to Image Synthesis using Stacked Conditional Variational Autoencoders and Conditional Generative Adversarial Networks ⬇️

Synthesizing a realistic image from textual description is a major challenge in computer vision. Current text to image synthesis approaches falls short of producing a highresolution image that represent a text descriptor. Most existing studies rely either on Generative Adversarial Networks (GANs) or Variational Auto Encoders (VAEs). GANs has the capability to produce sharper images but lacks the diversity of outputs, whereas VAEs are good at producing a diverse range of outputs, but the images generated are often blurred. Taking into account the relative advantages of both GANs and VAEs, we proposed a new stacked Conditional VAE (CVAE) and Conditional GAN (CGAN) network architecture for synthesizing images conditioned on a text description. This study uses Conditional VAEs as an initial generator to produce a high-level sketch of the text descriptor. This high-level sketch output from first stage and a text descriptor is used as an input to the conditional GAN network. The second stage GAN produces a 256x256 high resolution image. The proposed architecture benefits from a conditioning augmentation and a residual block on the Conditional GAN network to achieve the results. Multiple experiments were conducted using CUB and Oxford-102 dataset and the result of the proposed approach is compared against state-ofthe-art techniques such as StackGAN. The experiments illustrate that the proposed method generates a high-resolution image conditioned on text descriptions and yield competitive results based on Inception and Frechet Inception Score using both datasets

19.ExpansionNet: exploring the sequence length bottleneck in the Transformer for Image Captioning ⬇️

Most recent state of art architectures rely on combinations and variations of three approaches: convolutional, recurrent and self-attentive methods. Our work attempts in laying the basis for a new research direction for sequence modeling based upon the idea of modifying the sequence length. In order to do that, we propose a new method called ``Expansion Mechanism'' which transforms either dynamically or statically the input sequence into a new one featuring a different sequence length. Furthermore, we introduce a novel architecture that exploits such method and achieves competitive performances on the MS-COCO 2014 data set, yielding 134.6 and 131.4 CIDEr-D on the Karpathy test split in the ensemble and single model configuration respectively and 130 CIDEr-D in the official online testing server, despite being neither recurrent nor fully attentive. At the same time we address the efficiency aspect in our design and introduce a convenient training strategy suitable for most computational resources in contrast to the standard one. Source code is available at this https URL

20.D2HNet: Joint Denoising and Deblurring with Hierarchical Network for Robust Night Image Restoration ⬇️

Night imaging with modern smartphone cameras is troublesome due to low photon count and unavoidable noise in the imaging system. Directly adjusting exposure time and ISO ratings cannot obtain sharp and noise-free images at the same time in low-light conditions. Though many methods have been proposed to enhance noisy or blurry night images, their performances on real-world night photos are still unsatisfactory due to two main reasons: 1) Limited information in a single image and 2) Domain gap between synthetic training images and real-world photos (e.g., differences in blur area and resolution). To exploit the information from successive long- and short-exposure images, we propose a learning-based pipeline to fuse them. A D2HNet framework is developed to recover a high-quality image by deblurring and enhancing a long-exposure image under the guidance of a short-exposure image. To shrink the domain gap, we leverage a two-phase DeblurNet-EnhanceNet architecture, which performs accurate blur removal on a fixed low resolution so that it is able to handle large ranges of blur in different resolution inputs. In addition, we synthesize a D2-Dataset from HD videos and experiment on it. The results on the validation set and real photos demonstrate our methods achieve better visual quality and state-of-the-art quantitative scores. The D2HNet codes, models, and D2-Dataset can be found at this https URL.

21.Entropy-Based Feature Extraction For Real-Time Semantic Segmentation ⬇️

This paper introduces an efficient patch-based computational module, coined Entropy-based Patch Encoder (EPE) module, for resource-constrained semantic segmentation. The EPE module consists of three lightweight fully-convolutional encoders, each extracting features from image patches with a different amount of entropy. Patches with high entropy are being processed by the encoder with the largest number of parameters, patches with moderate entropy are processed by the encoder with a moderate number of parameters, and patches with low entropy are processed by the smallest encoder. The intuition behind the module is the following: as patches with high entropy contain more information, they need an encoder with more parameters, unlike low entropy patches, which can be processed using a small encoder. Consequently, processing part of the patches via the smaller encoder can significantly reduce the computational cost of the module. Experiments show that EPE can boost the performance of existing real-time semantic segmentation models with a slight increase in the computational cost. Specifically, EPE increases the mIOU performance of DFANet A by 0.9% with only 1.2% increase in the number of parameters and the mIOU performance of EDANet by 1% with 10% increase of the model parameters.

22.Dual Stream Computer-Generated Image Detection Network Based On Channel Joint And Softpool ⬇️

With the development of computer graphics technology, the images synthesized by computer software become more and more closer to the photographs. While computer graphics technology brings us a grand visual feast in the field of games and movies, it may also be utilized by someone with bad intentions to guide public opinions and cause political crisis or social unrest. Therefore, how to distinguish the computer-generated graphics (CG) from the photographs (PG) has become an important topic in the field of digital image forensics. This paper proposes a dual stream convolutional neural network based on channel joint and softpool. The proposed network architecture includes a residual module for extracting image noise information and a joint channel information extraction module for capturing the shallow semantic information of image. In addition, we also design a residual structure to enhance feature extraction and reduce the loss of information in residual flow. The joint channel information extraction module can obtain the shallow semantic information of the input image which can be used as the information supplement block of the residual module. The whole network uses SoftPool to reduce the information loss of down-sampling for image. Finally, we fuse the two flows to get the classification results. Experiments on SPL2018 and DsTok show that the proposed method outperforms existing methods, especially on the DsTok dataset. For example, the performance of our model surpasses the state-of-the-art by a large margin of 3%.

23.MCTS with Refinement for Proposals Selection Games in Scene Understanding ⬇️

We propose a novel method applicable in many scene understanding problems that adapts the Monte Carlo Tree Search (MCTS) algorithm, originally designed to learn to play games of high-state complexity. From a generated pool of proposals, our method jointly selects and optimizes proposals that minimize the objective term. In our first application for floor plan reconstruction from point clouds, our method selects and refines the room proposals, modelled as 2D polygons, by optimizing on an objective function combining the fitness as predicted by a deep network and regularizing terms on the room shapes. We also introduce a novel differentiable method for rendering the polygonal shapes of these proposals. Our evaluations on the recent and challenging Structured3D and Floor-SP datasets show significant improvements over the state-of-the-art, without imposing hard constraints nor assumptions on the floor plan configurations. In our second application, we extend our approach to reconstruct general 3D room layouts from a color image and obtain accurate room layouts. We also show that our differentiable renderer can easily be extended for rendering 3D planar polygons and polygon embeddings. Our method shows high performance on the Matterport3D-Layout dataset, without introducing hard constraints on room layout configurations.

24.FastHebb: Scaling Hebbian Training of Deep Neural Networks to ImageNet Level ⬇️

Learning algorithms for Deep Neural Networks are typically based on supervised end-to-end Stochastic Gradient Descent (SGD) training with error backpropagation (backprop). Backprop algorithms require a large number of labelled training samples to achieve high performance. However, in many realistic applications, even if there is plenty of image samples, very few of them are labelled, and semi-supervised sample-efficient training strategies have to be used. Hebbian learning represents a possible approach towards sample efficient training; however, in current solutions, it does not scale well to large datasets. In this paper, we present FastHebb, an efficient and scalable solution for Hebbian learning which achieves higher efficiency by 1) merging together update computation and aggregation over a batch of inputs, and 2) leveraging efficient matrix multiplication algorithms on GPU. We validate our approach on different computer vision benchmarks, in a semi-supervised learning scenario. FastHebb outperforms previous solutions by up to 50 times in terms of training speed, and notably, for the first time, we are able to bring Hebbian algorithms to ImageNet scale.

25.Self-Supervised Velocity Estimation for Automotive Radar Object Detection Networks ⬇️

This paper presents a method to learn the Cartesian velocity of objects using an object detection network on automotive radar data. The proposed method is self-supervised in terms of generating its own training signal for the velocities. Labels are only required for single-frame, oriented bounding boxes (OBBs). Labels for the Cartesian velocities or contiguous sequences, which are expensive to obtain, are not required. The general idea is to pre-train an object detection network without velocities using single-frame OBB labels, and then exploit the network's OBB predictions on unlabelled data for velocity training. In detail, the network's OBB predictions of the unlabelled frames are updated to the timestamp of a labelled frame using the predicted velocities and the distances between the updated OBBs of the unlabelled frame and the OBB predictions of the labelled frame are used to generate a self-supervised training signal for the velocities. The detection network architecture is extended by a module to account for the temporal relation of multiple scans and a module to represent the radars' radial velocity measurements explicitly. A two-step approach of first training only OBB detection, followed by training OBB detection and velocities is used. Further, a pre-training with pseudo-labels generated from radar radial velocity measurements bootstraps the self-supervised method of this paper. Experiments on the publicly available nuScenes dataset show that the proposed method almost reaches the velocity estimation performance of a fully supervised training, but does not require expensive velocity labels. Furthermore, we outperform a baseline method which uses only radial velocity measurements as labels.

26.Style Interleaved Learning for Generalizable Person Re-identification ⬇️

Domain generalization (DG) for person re-identification (ReID) is a challenging problem, as there is no access to target domain data permitted during the training process. Most existing DG ReID methods employ the same features for the updating of the feature extractor and classifier parameters. This common practice causes the model to overfit to existing feature styles in the source domain, resulting in sub-optimal generalization ability on target domains even if meta-learning is used. To solve this problem, we propose a novel style interleaved learning framework. Unlike conventional learning strategies, interleaved learning incorporates two forward propagations and one backward propagation for each iteration. We employ the features of interleaved styles to update the feature extractor and classifiers using different forward propagations, which helps the model avoid overfitting to certain domain styles. In order to fully explore the advantages of style interleaved learning, we further propose a novel feature stylization approach to diversify feature styles. This approach not only mixes the feature styles of multiple training samples, but also samples new and meaningful feature styles from batch-level style distribution. Extensive experimental results show that our model consistently outperforms state-of-the-art methods on large-scale benchmarks for DG ReID, yielding clear advantages in computational efficiency. Code is available at this https URL.

27.Multi-View Vision-to-Geometry Knowledge Transfer for 3D Point Cloud Shape Analysis ⬇️

As two fundamental representation modalities of 3D objects, 2D multi-view images and 3D point clouds reflect shape information from different aspects of visual appearances and geometric structures. Unlike deep learning-based 2D multi-view image modeling, which demonstrates leading performances in various 3D shape analysis tasks, 3D point cloud-based geometric modeling still suffers from insufficient learning capacity. In this paper, we innovatively construct a unified cross-modal knowledge transfer framework, which distills discriminative visual descriptors of 2D images into geometric descriptors of 3D point clouds. Technically, under a classic teacher-student learning paradigm, we propose multi-view vision-to-geometry distillation, consisting of a deep 2D image encoder as teacher and a deep 3D point cloud encoder as student. To achieve heterogeneous feature alignment, we further propose visibility-aware feature projection, through which per-point embeddings can be aggregated into multi-view geometric descriptors. Extensive experiments on 3D shape classification, part segmentation, and unsupervised learning validate the superiority of our method. We will make the code and data publicly available.

28.Design of Human Machine Interface through vision-based low-cost Hand Gesture Recognition system based on deep CNN with transfer-learning approach ⬇️

In this work, a real-time hand gesture recognition system-based human-computer interface (HCI) is presented. The system consists of six stages: (1) hand detection, (2) gesture segmentation, (3) use of six pre-trained CNN models by using the transfer-learning method, (4) building an interactive human-machine interface, (5) development of a gesture-controlled virtual mouse, (6) use of Kalman filter to estimate the hand position, based on that the smoothness of the motion of pointer is improved. Six pre-trained convolutional neural network (CNN) models (VGG16, VGG19, ResNet50, ResNet101, Inception-V1, and MobileNet-V1) have been used to classify hand gesture images. Three multi-class datasets (two publicly and one custom) have been used to evaluate the model performances. Considering the models' performances, it has been observed that Inception-V1 has significantly shown a better classification performance compared to the other five pre-trained models in terms of accuracy, precision, recall, and F-score values. The gesture recognition system is expanded and used to control multimedia applications (like VLC player, audio player, file management, playing 2D Super-Mario-Bros game, etc.) with different customized gesture commands in real-time scenarios. The average speed of this system has reached 35 fps (frame per seconds), which meets the requirements for the real-time scenario.

29.Masked Surfel Prediction for Self-Supervised Point Cloud Learning ⬇️

Masked auto-encoding is a popular and effective self-supervised learning approach to point cloud learning. However, most of the existing methods reconstruct only the masked points and overlook the local geometry information, which is also important to understand the point cloud data. In this work, we make the first attempt, to the best of our knowledge, to consider the local geometry information explicitly into the masked auto-encoding, and propose a novel Masked Surfel Prediction (MaskSurf) method. Specifically, given the input point cloud masked at a high ratio, we learn a transformer-based encoder-decoder network to estimate the underlying masked surfels by simultaneously predicting the surfel positions (i.e., points) and per-surfel orientations (i.e., normals). The predictions of points and normals are supervised by the Chamfer Distance and a newly introduced Position-Indexed Normal Distance in a set-to-set manner. Our MaskSurf is validated on six downstream tasks under three fine-tuning strategies. In particular, MaskSurf outperforms its closest competitor, Point-MAE, by 1.2% on the real-world dataset of ScanObjectNN under the OBJ-BG setting, justifying the advantages of masked surfel prediction over masked point cloud reconstruction. Codes will be available at this https URL.

30.EPIC-KITCHENS-100 Unsupervised Domain Adaptation Challenge for Action Recognition 2022: Team HNU-FPV Technical Report ⬇️

In this report, we present the technical details of our submission to the 2022 EPIC-Kitchens Unsupervised Domain Adaptation (UDA) Challenge. Existing UDA methods align the global features extracted from the whole video clips across the source and target domains but suffer from the spatial redundancy of feature matching in video recognition. Motivated by the observation that in most cases a small image region in each video frame can be informative enough for the action recognition task, we propose to exploit informative image regions to perform efficient domain alignment. Specifically, we first use lightweight CNNs to extract the global information of the input two-stream video frames and select the informative image patches by a differentiable interpolation-based selection strategy. Then the global information from videos frames and local information from image patches are processed by an existing video adaptation method, i.e., TA3N, in order to perform feature alignment for the source domain and the target domain. Our method (without model ensemble) ranks 4th among this year's teams on the test set of EPIC-KITCHENS-100.

31.Adaptation of Surgical Activity Recognition Models Across Operating Rooms ⬇️

Automatic surgical activity recognition enables more intelligent surgical devices and a more efficient workflow. Integration of such technology in new operating rooms has the potential to improve care delivery to patients and decrease costs. Recent works have achieved a promising performance on surgical activity recognition; however, the lack of generalizability of these models is one of the critical barriers to the wide-scale adoption of this technology. In this work, we study the generalizability of surgical activity recognition models across operating rooms. We propose a new domain adaptation method to improve the performance of the surgical activity recognition model in a new operating room for which we only have unlabeled videos. Our approach generates pseudo labels for unlabeled video clips that it is confident about and trains the model on the augmented version of the clips. We extend our method to a semi-supervised domain adaptation setting where a small portion of the target domain is also labeled. In our experiments, our proposed method consistently outperforms the baselines on a dataset of more than 480 long surgical videos collected from two operating rooms.

32.DRL-ISP: Multi-Objective Camera ISP with Deep Reinforcement Learning ⬇️

In this paper, we propose a multi-objective camera ISP framework that utilizes Deep Reinforcement Learning (DRL) and camera ISP toolbox that consist of network-based and conventional ISP tools. The proposed DRL-based camera ISP framework iteratively selects a proper tool from the toolbox and applies it to the image to maximize a given vision task-specific reward function. For this purpose, we implement total 51 ISP tools that include exposure correction, color-and-tone correction, white balance, sharpening, denoising, and the others. We also propose an efficient DRL network architecture that can extract the various aspects of an image and make a rigid mapping relationship between images and a large number of actions. Our proposed DRL-based ISP framework effectively improves the image quality according to each vision task such as RAW-to-RGB image restoration, 2D object detection, and monocular depth estimation.

33.Contrastive Learning from Spatio-Temporal Mixed Skeleton Sequences for Self-Supervised Skeleton-Based Action Recognition ⬇️

Self-supervised skeleton-based action recognition with contrastive learning has attracted much attention. Recent literature shows that data augmentation and large sets of contrastive pairs are crucial in learning such representations. In this paper, we found that directly extending contrastive pairs based on normal augmentations brings limited returns in terms of performance, because the contribution of contrastive pairs from the normal data augmentation to the loss get smaller as training progresses. Therefore, we delve into hard contrastive pairs for contrastive learning. Motivated by the success of mixing augmentation strategy which improves the performance of many tasks by synthesizing novel samples, we propose SkeleMixCLR: a contrastive learning framework with a spatio-temporal skeleton mixing augmentation (SkeleMix) to complement current contrastive learning approaches by providing hard contrastive samples. First, SkeleMix utilizes the topological information of skeleton data to mix two skeleton sequences by randomly combing the cropped skeleton fragments (the trimmed view) with the remaining skeleton sequences (the truncated view). Second, a spatio-temporal mask pooling is applied to separate these two views at the feature level. Third, we extend contrastive pairs with these two views. SkeleMixCLR leverages the trimmed and truncated views to provide abundant hard contrastive pairs since they involve some context information from each other due to the graph convolution operations, which allows the model to learn better motion representations for action recognition. Extensive experiments on NTU-RGB+D, NTU120-RGB+D, and PKU-MMD datasets show that SkeleMixCLR achieves state-of-the-art performance. Codes are available at this https URL.

34.Shadow-Background-Noise 3D Spatial Decomposition Using Sparse Low-Rank Gaussian Properties for Video-SAR Moving Target Shadow Enhancement ⬇️

Moving target shadows among video synthetic aperture radar (Video-SAR) images are always interfered by low scattering backgrounds and cluttered noises, causing poor moving target shadow detection-tracking performance. To solve this problem, this letter proposes a shadow-background-noise 3D spatial de-composition method named SBN-3D-SD to boost shadow saliency for better Video-SAR moving target shadow detection-tracking performance.

35.Deep Rotation Correction without Angle Prior ⬇️

Not everybody can be equipped with professional photography skills and sufficient shooting time, and there can be some tilts in the captured images occasionally. In this paper, we propose a new and practical task, named Rotation Correction, to automatically correct the tilt with high content fidelity in the condition that the rotated angle is unknown. This task can be easily integrated into image editing applications, allowing users to correct the rotated images without any manual operations. To this end, we leverage a neural network to predict the optical flows that can warp the tilted images to be perceptually horizontal. Nevertheless, the pixel-wise optical flow estimation from a single image is severely unstable, especially in large-angle tilted images. To enhance its robustness, we propose a simple but effective prediction strategy to form a robust elastic warp. Particularly, we first regress the mesh deformation that can be transformed into robust initial optical flows. Then we estimate residual optical flows to facilitate our network the flexibility of pixel-wise deformation, further correcting the details of the tilted images. To establish an evaluation benchmark and train the learning framework, a comprehensive rotation correction dataset is presented with a large diversity in scenes and rotated angles. Extensive experiments demonstrate that even in the absence of the angle prior, our algorithm can outperform other state-of-the-art solutions requiring this prior. The codes and dataset will be available at this https URL.

36.AV-Gaze: A Study on the Effectiveness of Audio Guided Visual Attention Estimation for Non-Profilic Faces ⬇️

In challenging real-life conditions such as extreme head-pose, occlusions, and low-resolution images where the visual information fails to estimate visual attention/gaze direction, audio signals could provide important and complementary information. In this paper, we explore if audio-guided coarse head-pose can further enhance visual attention estimation performance for non-prolific faces. Since it is difficult to annotate audio signals for estimating the head-pose of the speaker, we use off-the-shelf state-of-the-art models to facilitate cross-modal weak-supervision. During the training phase, the framework learns complementary information from synchronized audio-visual modality. Our model can utilize any of the available modalities i.e. audio, visual or audio-visual for task-specific inference. It is interesting to note that, when AV-Gaze is tested on benchmark datasets with these specific modalities, it achieves competitive results on multiple datasets, while being highly adaptive towards challenging scenarios.

37.Single-image Defocus Deblurring by Integration of Defocus Map Prediction Tracing the Inverse Problem Computation ⬇️

In this paper, we consider the problem in defocus image deblurring. Previous classical methods follow two-steps approaches, i.e., first defocus map estimation and then the non-blind deblurring. In the era of deep learning, some researchers have tried to address these two problems by CNN. However, the simple concatenation of defocus map, which represents the blur level, leads to suboptimal performance. Considering the spatial variant property of the defocus blur and the blur level indicated in the defocus map, we employ the defocus map as conditional guidance to adjust the features from the input blurring images instead of simple concatenation. Then we propose a simple but effective network with spatial modulation based on the defocus map. To achieve this, we design a network consisting of three sub-networks, including the defocus map estimation network, a condition network that encodes the defocus map into condition features, and the defocus deblurring network that performs spatially dynamic modulation based on the condition features. Moreover, the spatially dynamic modulation is based on an affine transform function to adjust the features from the input blurry images. Experimental results show that our method can achieve better quantitative and qualitative evaluation performance than the existing state-of-the-art methods on the commonly used public test datasets.

38.Vision Transformers: State of the Art and Research Challenges ⬇️

Transformers have achieved great success in natural language processing. Due to the powerful capability of self-attention mechanism in transformers, researchers develop the vision transformers for a variety of computer vision tasks, such as image recognition, object detection, image segmentation, pose estimation, and 3D reconstruction. This paper presents a comprehensive overview of the literature on different architecture designs and training tricks (including self-supervised learning) for vision transformers. Our goal is to provide a systematic review with the open research opportunities.

39.Dual-Stream Transformer for Generic Event Boundary Captioning ⬇️

This paper describes our champion solution for the CVPR2022 Generic Event Boundary Captioning (GEBC) competition. GEBC requires the captioning model to have a comprehension of instantaneous status changes around the given video boundary, which makes it much more challenging than conventional video captioning task. In this paper, a Dual-Stream Transformer with improvements on both video content encoding and captions generation is proposed: (1) We utilize three pre-trained models to extract the video features from different granularities. Moreover, we exploit the types of boundary as hints to help the model generate captions. (2) We particularly design an model, termed as Dual-Stream Transformer, to learn discriminative representations for boundary captioning. (3) Towards generating content-relevant and human-like captions, we improve the description quality by designing a word-level ensemble strategy. The promising results on the GEBC test split demonstrate the efficacy of our proposed model.

40.Partial Shape Similarity via Alignment of Multi-Metric Hamiltonian Spectra ⬇️

Evaluating the similarity of non-rigid shapes with significant partiality is a fundamental task in numerous computer vision applications. Here, we propose a novel axiomatic method to match similar regions across shapes. Matching similar regions is formulated as the alignment of the spectra of operators closely related to the Laplace-Beltrami operator (LBO). The main novelty of the proposed approach is the consideration of differential operators defined on a manifold with multiple metrics. The choice of a metric relates to fundamental shape properties while considering the same manifold under different metrics can thus be viewed as analyzing the underlying manifold from different perspectives. Specifically, we examine the scale-invariant metric and the corresponding scale-invariant Laplace-Beltrami operator (SI-LBO) along with the regular metric and the regular LBO. We demonstrate that the scale-invariant metric emphasizes the locations of important semantic features in articulated shapes. A truncated spectrum of the SI-LBO consequently better captures locally curved regions and complements the global information encapsulated in the truncated spectrum of the regular LBO. We show that matching these dual spectra outperforms competing axiomatic frameworks when tested on standard benchmarks. We introduced a new dataset and compare the proposed method with the state-of-the-art learning based approach in a cross-database configuration. Specifically, we show that, when trained on one data set and tested on another, the proposed axiomatic approach which does not involve training, outperforms the deep learning alternative.

41.MaiT: Leverage Attention Masks for More Efficient Image Transformers ⬇️

Though image transformers have shown competitive results with convolutional neural networks in computer vision tasks, lacking inductive biases such as locality still poses problems in terms of model efficiency especially for embedded applications. In this work, we address this issue by introducing attention masks to incorporate spatial locality into self-attention heads. Local dependencies are captured efficiently with masked attention heads along with global dependencies captured by unmasked attention heads. With Masked attention image Transformer - MaiT, top-1 accuracy increases by up to 1.7% compared to CaiT with fewer parameters and FLOPs, and the throughput improves by up to 1.5X compared to Swin. Encoding locality with attention masks is model agnostic, and thus it applies to monolithic, hierarchical, or other novel transformer architectures.

42.Semi-supervised Human Pose Estimation in Art-historical Images ⬇️

Gesture as \enquote*{language} of non-verbal communication has been theoretically established since the 17th century. However, its relevance for the visual arts has been expressed only sporadically. This may be primarily due to the sheer overwhelming amount of data that traditionally had to be processed by hand. With the steady progress of digitization, though, a growing number of historical artifacts have been indexed and made available to the public, creating a need for automatic retrieval of art-historical motifs with similar body constellations or poses. Since the domain of art differs significantly from existing real-world data sets for human pose estimation due to its style variance, this presents new challenges. In this paper, we propose a novel approach to estimate human poses in art-historical images. In contrast to previous work that attempts to bridge the domain gap with pre-trained models or through style transfer, we suggest semi-supervised learning for both object and keypoint detection. Furthermore, we introduce a novel domain-specific art data set that includes both bounding box and keypoint annotations of human figures. Our approach achieves significantly better results than methods that use pre-trained models or style transfer.

43.Network Binarization via Contrastive Learning ⬇️

Neural network binarization accelerates deep models by quantizing their weights and activations into 1-bit. However, there is still a huge performance gap between Binary Neural Networks (BNNs) and their full-precision (FP) counterparts. As the quantization error caused by weights binarization has been reduced in earlier works, the activations binarization becomes the major obstacle for further improvement of the accuracy. BNN characterises a unique and interesting structure, where the binary and latent FP activations exist in the same forward pass (\textit{i.e.} $\text{Binarize}(\mathbf{a}_F) = \mathbf{a}_B$). To mitigate the information degradation caused by the binarization operation from FP to binary activations, we establish a novel contrastive learning framework while training BNNs through the lens of Mutual Information (MI) maximization. MI is introduced as the metric to measure the information shared between binary and FP activations, which assists binarization with contrastive learning. Specifically, the representation ability of the BNNs is greatly strengthened via pulling the positive pairs with binary and FP activations from the same input samples, as well as pushing negative pairs from different samples (the number of negative pairs can be exponentially large). This benefits the downstream tasks, not only classification but also segmentation and depth estimation,~\textit{etc}. The experimental results show that our method can be implemented as a pile-up module on existing state-of-the-art binarization methods and can remarkably improve the performance over them on CIFAR-10/100 and ImageNet, in addition to the great generalization ability on NYUD-v2.

44.The Weaknesses of Adversarial Camouflage in Overhead Imagery ⬇️

Machine learning is increasingly critical for analysis of the ever-growing corpora of overhead imagery. Advanced computer vision object detection techniques have demonstrated great success in identifying objects of interest such as ships, automobiles, and aircraft from satellite and drone imagery. Yet relying on computer vision opens up significant vulnerabilities, namely, the susceptibility of object detection algorithms to adversarial attacks. In this paper we explore the efficacy and drawbacks of adversarial camouflage in an overhead imagery context. While a number of recent papers have demonstrated the ability to reliably fool deep learning classifiers and object detectors with adversarial patches, most of this work has been performed on relatively uniform datasets and only a single class of objects. In this work we utilize the VisDrone dataset, which has a large range of perspectives and object sizes. We explore four different object classes: bus, car, truck, van. We build a library of 24 adversarial patches to disguise these objects, and introduce a patch translucency variable to our patches. The translucency (or alpha value) of the patches is highly correlated to their efficacy. Further, we show that while adversarial patches may fool object detectors, the presence of such patches is often easily uncovered, with patches on average 24% more detectable than the objects the patches were meant to hide. This raises the question of whether such patches truly constitute camouflage. Source code is available at this https URL.

45.SphereVLAD++: Attention-based and Signal-enhanced Viewpoint Invariant Descriptor ⬇️

LiDAR-based localization approach is a fundamental module for large-scale navigation tasks, such as last-mile delivery and autonomous driving, and localization robustness highly relies on viewpoints and 3D feature extraction. Our previous work provides a viewpoint-invariant descriptor to deal with viewpoint differences; however, the global descriptor suffers from a low signal-noise ratio in unsupervised clustering, reducing the distinguishable feature extraction ability. We develop SphereVLAD++, an attention-enhanced viewpoint invariant place recognition method in this work. SphereVLAD++ projects the point cloud on the spherical perspective for each unique area and captures the contextual connections between local features and their dependencies with global 3D geometry distribution. In return, clustered elements within the global descriptor are conditioned on local and global geometries and support the original viewpoint-invariant property of SphereVLAD. In the experiments, we evaluated the localization performance of SphereVLAD++ on both public KITTI360 datasets and self-generated datasets from the city of Pittsburgh. The experiment results show that SphereVLAD++ outperforms all relative state-of-the-art 3D place recognition methods under small or even totally reversed viewpoint differences and shows 0.69% and 15.81% successful retrieval rates with better than the second best. Low computation requirements and high time efficiency also help its application for low-cost robots.

46.Towards Transparency in Dermatology Image Datasets with Skin Tone Annotations by Experts, Crowds, and an Algorithm ⬇️

While artificial intelligence (AI) holds promise for supporting healthcare providers and improving the accuracy of medical diagnoses, a lack of transparency in the composition of datasets exposes AI models to the possibility of unintentional and avoidable mistakes. In particular, public and private image datasets of dermatological conditions rarely include information on skin color. As a start towards increasing transparency, AI researchers have appropriated the use of the Fitzpatrick skin type (FST) from a measure of patient photosensitivity to a measure for estimating skin tone in algorithmic audits of computer vision applications including facial recognition and dermatology diagnosis. In order to understand the variability of estimated FST annotations on images, we compare several FST annotation methods on a diverse set of 460 images of skin conditions from both textbooks and online dermatology atlases. We find the inter-rater reliability between three board-certified dermatologists is comparable to the inter-rater reliability between the board-certified dermatologists and two crowdsourcing methods. In contrast, we find that the Individual Typology Angle converted to FST (ITA-FST) method produces annotations that are significantly less correlated with the experts' annotations than the experts' annotations are correlated with each other. These results demonstrate that algorithms based on ITA-FST are not reliable for annotating large-scale image datasets, but human-centered, crowd-based protocols can reliably add skin type transparency to dermatology datasets. Furthermore, we introduce the concept of dynamic consensus protocols with tunable parameters including expert review that increase the visibility of crowdwork and provide guidance for future crowdsourced annotations of large image datasets.

47.Efficient fine-grained road segmentation using superpixel-based CNN and CRF models ⬇️

Towards a safe and comfortable driving, road scene segmentation is a rudimentary problem in camera-based advance driver assistance systems (ADAS). Despite of the great achievement of Convolutional Neural Networks (CNN) for semantic segmentation task, the high computational efforts of CNN based methods is still a challenging area. In recent work, we proposed a novel approach to utilise the advantages of CNNs for the task of road segmentation at reasonable computational effort. The runtime benefits from using irregular super pixels as basis for the input for the CNN rather than the image grid, which tremendously reduces the input size. Although, this method achieved remarkable low computational time in both training and testing phases, the lower resolution of the super pixel domain yields naturally lower accuracy compared to high cost state of the art methods. In this work, we focus on a refinement of the road segmentation utilising a Conditional Random Field (CRF).The refinement procedure is limited to the super pixels touching the predicted road boundary to keep the additional computational effort low. Reducing the input to the super pixel domain allows the CNNs structure to stay small and efficient to compute while keeping the advantage of convolutional layers and makes them eligible for ADAS. Applying CRF compensate the trade off between accuracy and computational efficiency. The proposed system obtained comparable performance among the top performing algorithms on the KITTI road benchmark and its fast inference makes it particularly suitable for realtime applications.

48.TFCNs: A CNN-Transformer Hybrid Network for Medical Image Segmentation ⬇️

Medical image segmentation is one of the most fundamental tasks concerning medical information analysis. Various solutions have been proposed so far, including many deep learning-based techniques, such as U-Net, FC-DenseNet, etc. However, high-precision medical image segmentation remains a highly challenging task due to the existence of inherent magnification and distortion in medical images as well as the presence of lesions with similar density to normal tissues. In this paper, we propose TFCNs (Transformers for Fully Convolutional denseNets) to tackle the problem by introducing ResLinear-Transformer (RL-Transformer) and Convolutional Linear Attention Block (CLAB) to FC-DenseNet. TFCNs is not only able to utilize more latent information from the CT images for feature extraction, but also can capture and disseminate semantic features and filter non-semantic features more effectively through the CLAB module. Our experimental results show that TFCNs can achieve state-of-the-art performance with dice scores of 83.72% on the Synapse dataset. In addition, we evaluate the robustness of TFCNs for lesion area effects on the COVID-19 public datasets. The Python code will be made publicly available on this https URL.

49.Learning to restore images degraded by atmospheric turbulence using uncertainty ⬇️

Atmospheric turbulence can significantly degrade the quality of images acquired by long-range imaging systems by causing spatially and temporally random fluctuations in the index of refraction of the atmosphere. Variations in the refractive index causes the captured images to be geometrically distorted and blurry. Hence, it is important to compensate for the visual degradation in images caused by atmospheric turbulence. In this paper, we propose a deep learning-based approach for restring a single image degraded by atmospheric turbulence. We make use of the epistemic uncertainty based on Monte Carlo dropouts to capture regions in the image where the network is having hard time restoring. The estimated uncertainty maps are then used to guide the network to obtain the restored image. Extensive experiments are conducted on synthetic and real images to show the significance of the proposed work. Code is available at : this https URL

50.Fairness and Bias in Robot Learning ⬇️

Machine learning has significantly enhanced the abilities of robots, enabling them to perform a wide range of tasks in human environments and adapt to our uncertain real world. Recent works in various domains of machine learning have highlighted the importance of accounting for fairness to ensure that these algorithms do not reproduce human biases and consequently lead to discriminatory outcomes. With robot learning systems increasingly performing more and more tasks in our everyday lives, it is crucial to understand the influence of such biases to prevent unintended behavior toward certain groups of people. In this work, we present the first survey on fairness in robot learning from an interdisciplinary perspective spanning technical, ethical, and legal challenges. We propose a taxonomy for sources of bias and the resulting types of discrimination due to them. Using examples from different robot learning domains, we examine scenarios of unfair outcomes and strategies to mitigate them. We present early advances in the field by covering different fairness definitions, ethical and legal considerations, and methods for fair robot learning. With this work, we aim at paving the road for groundbreaking developments in fair robot learning.

51.Back to the Source: Diffusion-Driven Test-Time Adaptation ⬇️

Test-time adaptation harnesses test inputs to improve the accuracy of a model trained on source data when tested on shifted target data. Existing methods update the source model by (re-)training on each target domain. While effective, re-training is sensitive to the amount and order of the data and the hyperparameters for optimization. We instead update the target data, by projecting all test inputs toward the source domain with a generative diffusion model. Our diffusion-driven adaptation method, DDA, shares its models for classification and generation across all domains. Both models are trained on the source domain, then fixed during testing. We augment diffusion with image guidance and self-ensembling to automatically decide how much to adapt. Input adaptation by DDA is more robust than prior model adaptation approaches across a variety of corruptions, architectures, and data regimes on the ImageNet-C benchmark. With its input-wise updates, DDA succeeds where model adaptation degrades on too little data in small batches, dependent data in non-uniform order, or mixed data with multiple corruptions.

52.A Novel Unified Conditional Score-based Generative Framework for Multi-modal Medical Image Completion ⬇️

Multi-modal medical image completion has been extensively applied to alleviate the missing modality issue in a wealth of multi-modal diagnostic tasks. However, for most existing synthesis methods, their inferences of missing modalities can collapse into a deterministic mapping from the available ones, ignoring the uncertainties inherent in the cross-modal relationships. Here, we propose the Unified Multi-Modal Conditional Score-based Generative Model (UMM-CSGM) to take advantage of Score-based Generative Model (SGM) in modeling and stochastically sampling a target probability distribution, and further extend SGM to cross-modal conditional synthesis for various missing-modality configurations in a unified framework. Specifically, UMM-CSGM employs a novel multi-in multi-out Conditional Score Network (mm-CSN) to learn a comprehensive set of cross-modal conditional distributions via conditional diffusion and reverse generation in the complete modality space. In this way, the generation process can be accurately conditioned by all available information, and can fit all possible configurations of missing modalities in a single network. Experiments on BraTS19 dataset show that the UMM-CSGM can more reliably synthesize the heterogeneous enhancement and irregular area in tumor-induced lesions for any missing modalities.

53.BMD-GAN: Bone mineral density estimation using x-ray image decomposition into projections of bone-segmented quantitative computed tomography using hierarchical learning ⬇️

We propose a method for estimating the bone mineral density (BMD) from a plain x-ray image. Dual-energy X-ray absorptiometry (DXA) and quantitative computed tomography (QCT) provide high accuracy in diagnosing osteoporosis; however, these modalities require special equipment and scan protocols. Measuring BMD from an x-ray image provides an opportunistic screening, which is potentially useful for early diagnosis. The previous methods that directly learn the relationship between x-ray images and BMD require a large training dataset to achieve high accuracy because of large intensity variations in the x-ray images. Therefore, we propose an approach using the QCT for training a generative adversarial network (GAN) and decomposing an x-ray image into a projection of bone-segmented QCT. The proposed hierarchical learning improved the robustness and accuracy of quantitatively decomposing a small-area target. The evaluation of 200 patients with osteoarthritis using the proposed method, which we named BMD-GAN, demonstrated a Pearson correlation coefficient of 0.888 between the predicted and ground truth DXA-measured BMD. Besides not requiring a large-scale training database, another advantage of our method is its extensibility to other anatomical areas, such as the vertebrae and rib bones.

54.Self-Supervised Learning of Music-Dance Representation through Explicit-Implicit Rhythm Synchronization ⬇️

Although audio-visual representation has been proved to be applicable in many downstream tasks, the representation of dancing videos, which is more specific and always accompanied by music with complex auditory contents, remains challenging and uninvestigated. Considering the intrinsic alignment between the cadent movement of dancer and music rhythm, we introduce MuDaR, a novel Music-Dance Representation learning framework to perform the synchronization of music and dance rhythms both in explicit and implicit ways. Specifically, we derive the dance rhythms based on visual appearance and motion cues inspired by the music rhythm analysis. Then the visual rhythms are temporally aligned with the music counterparts, which are extracted by the amplitude of sound intensity. Meanwhile, we exploit the implicit coherence of rhythms implied in audio and visual streams by contrastive learning. The model learns the joint embedding by predicting the temporal consistency between audio-visual pairs. The music-dance representation, together with the capability of detecting audio and visual rhythms, can further be applied to three downstream tasks: (a) dance classification, (b) music-dance retrieval, and (c) music-dance retargeting. Extensive experiments demonstrate that our proposed framework outperforms other self-supervised methods by a large margin.

55.Uncertainty of Atmospheric Motion Vectors by Sampling Tempered Posterior Distributions ⬇️

Atmospheric motion vectors (AMVs) extracted from satellite imagery are the only wind observations with good global coverage. They are important features for feeding numerical weather prediction (NWP) models. Several Bayesian models have been proposed to estimate AMVs. Although critical for correct assimilation into NWP models, very few methods provide a thorough characterization of the estimation errors. The difficulty of estimating errors stems from the specificity of the posterior distribution, which is both very high dimensional, and highly ill-conditioned due to a singular likelihood, which becomes critical in particular in the case of missing data (unobserved pixels). This work studies the evaluation of the expected error of AMVs using gradient-based Markov Chain Monte Carlo (MCMC) algorithms. Our main contribution is to propose a tempering strategy, which amounts to sampling a local approximation of the joint posterior distribution of AMVs and image variables in the neighborhood of a point estimate. In addition, we provide efficient preconditioning with the covariance related to the prior family itself (fractional Brownian motion), with possibly different hyper-parameters. From a theoretical point of view, we show that under regularity assumptions, the family of tempered posterior distributions converges in distribution as temperature decreases to an {optimal} Gaussian approximation at a point estimate given by the Maximum A Posteriori (MAP) log-density. From an empirical perspective, we evaluate the proposed approach based on some quantitative Bayesian evaluation criteria. Our numerical simulations performed on synthetic and real meteorological data reveal a significant gain in terms of accuracy of the AMV point estimates and of their associated expected error estimates, but also a substantial acceleration in the convergence speed of the MCMC algorithms.

56.Deformer: Towards Displacement Field Learning for Unsupervised Medical Image Registration ⬇️

Recently, deep-learning-based approaches have been widely studied for deformable image registration task. However, most efforts directly map the composite image representation to spatial transformation through the convolutional neural network, ignoring its limited ability to capture spatial correspondence. On the other hand, Transformer can better characterize the spatial relationship with attention mechanism, its long-range dependency may be harmful to the registration task, where voxels with too large distances are unlikely to be corresponding pairs. In this study, we propose a novel Deformer module along with a multi-scale framework for the deformable image registration task. The Deformer module is designed to facilitate the mapping from image representation to spatial transformation by formulating the displacement vector prediction as the weighted summation of several bases. With the multi-scale framework to predict the displacement fields in a coarse-to-fine manner, superior performance can be achieved compared with traditional and learning-based approaches. Comprehensive experiments on two public datasets are conducted to demonstrate the effectiveness of the proposed Deformer module as well as the multi-scale framework.

57.Improving Few-Shot Image Classification Using Machine- and User-Generated Natural Language Descriptions ⬇️

Humans can obtain the knowledge of novel visual concepts from language descriptions, and we thus use the few-shot image classification task to investigate whether a machine learning model can have this capability. Our proposed model, LIDE (Learning from Image and DEscription), has a text decoder to generate the descriptions and a text encoder to obtain the text representations of machine- or user-generated descriptions. We confirmed that LIDE with machine-generated descriptions outperformed baseline models. Moreover, the performance was improved further with high-quality user-generated descriptions. The generated descriptions can be viewed as the explanations of the model's predictions, and we observed that such explanations were consistent with prediction results. We also investigated why the language description improved the few-shot image classification performance by comparing the image representations and the text representations in the feature spaces.

58.Uncertainty-Aware Self-supervised Neural Network for Liver $T_{1ρ}$ Mapping with Relaxation Constraint ⬇️

$T_{1\rho}$ mapping is a promising quantitative MRI technique for the non-invasive assessment of tissue properties. Learning-based approaches can map $T_{1\rho}$ from a reduced number of $T_{1\rho}$ weighted images, but requires significant amounts of high quality training data. Moreover, existing methods do not provide the confidence level of the $T_{1\rho}$ estimation. To address these problems, we proposed a self-supervised learning neural network that learns a $T_{1\rho}$ mapping using the relaxation constraint in the learning process. Epistemic uncertainty and aleatoric uncertainty are modelled for the $T_{1\rho}$ quantification network to provide a Bayesian confidence estimation of the $T_{1\rho}$ mapping. The uncertainty estimation can also regularize the model to prevent it from learning imperfect data. We conducted experiments on $T_{1\rho}$ data collected from 52 patients with non-alcoholic fatty liver disease. The results showed that our method outperformed the existing methods for $T_{1\rho}$ quantification of the liver using as few as two $T_{1\rho}$-weighted images. Our uncertainty estimation provided a feasible way of modelling the confidence of the self-supervised learning based $T_{1\rho}$ estimation, which is consistent with the reality in liver $T_{1\rho}$ imaging.

59.What Makes for Automatic Reconstruction of Pulmonary Segments ⬇️

3D reconstruction of pulmonary segments plays an important role in surgical treatment planning of lung cancer, which facilitates preservation of pulmonary function and helps ensure low recurrence rates. However, automatic reconstruction of pulmonary segments remains unexplored in the era of deep learning. In this paper, we investigate what makes for automatic reconstruction of pulmonary segments. First and foremost, we formulate, clinically and geometrically, the anatomical definitions of pulmonary segments, and propose evaluation metrics adhering to these definitions. Second, we propose ImPulSe (Implicit Pulmonary Segment), a deep implicit surface model designed for pulmonary segment reconstruction. The automatic reconstruction of pulmonary segments by ImPulSe is accurate in metrics and visually appealing. Compared with canonical segmentation methods, ImPulSe outputs continuous predictions of arbitrary resolutions with higher training efficiency and fewer parameters. Lastly, we experiment with different network inputs to analyze what matters in the task of pulmonary segment reconstruction. Our code is available at this https URL.

60.Back to the Basics: Revisiting Out-of-Distribution Detection Baselines ⬇️

We study simple methods for out-of-distribution (OOD) image detection that are compatible with any already trained classifier, relying on only its predictions or learned representations. Evaluating the OOD detection performance of various methods when utilized with ResNet-50 and Swin Transformer models, we find methods that solely consider the model's predictions can be easily outperformed by also considering the learned representations. Based on our analysis, we advocate for a dead-simple approach that has been neglected in other studies: simply flag as OOD images whose average distance to their K nearest neighbors is large (in the representation space of an image classifier trained on the in-distribution data).

61.Multi-Task Lung Nodule Detection in Chest Radiographs with a Dual Head Network ⬇️

Lung nodules can be an alarming precursor to potential lung cancer. Missed nodule detections during chest radiograph analysis remains a common challenge among thoracic radiologists. In this work, we present a multi-task lung nodule detection algorithm for chest radiograph analysis. Unlike past approaches, our algorithm predicts a global-level label indicating nodule presence along with local-level labels predicting nodule locations using a Dual Head Network (DHN). We demonstrate the favorable nodule detection performance that our multi-task formulation yields in comparison to conventional methods. In addition, we introduce a novel Dual Head Augmentation (DHA) strategy tailored for DHN, and we demonstrate its significance in further enhancing global and local nodule predictions.

62.Self-Supervised RF Signal Representation Learning for NextG Signal Classification with Deep Learning ⬇️

Deep learning (DL) finds rich applications in the wireless domain to improve spectrum awareness. Typically, the DL models are either randomly initialized following a statistical distribution or pretrained on tasks from other data domains such as computer vision (in the form of transfer learning) without accounting for the unique characteristics of wireless signals. Self-supervised learning enables the learning of useful representations from Radio Frequency (RF) signals themselves even when only limited training data samples with labels are available. We present the first self-supervised RF signal representation learning model and apply it to the automatic modulation recognition (AMR) task by specifically formulating a set of transformations to capture the wireless signal characteristics. We show that the sample efficiency (the number of labeled samples required to achieve a certain accuracy performance) of AMR can be significantly increased (almost an order of magnitude) by learning signal representations with self-supervised learning. This translates to substantial time and cost savings. Furthermore, self-supervised learning increases the model accuracy compared to the state-of-the-art DL methods and maintains high accuracy even when a small set of training data samples is used.

63.Orthogonal Matrix Retrieval with Spatial Consensus for 3D Unknown-View Tomography ⬇️

Unknown-view tomography (UVT) reconstructs a 3D density map from its 2D projections at unknown, random orientations. A line of work starting with Kam (1980) employs the method of moments (MoM) with rotation-invariant Fourier features to solve UVT in the frequency domain, assuming that the orientations are uniformly distributed. This line of work includes the recent orthogonal matrix retrieval (OMR) approaches based on matrix factorization, which, while elegant, either require side information about the density that is not available, or fail to be sufficiently robust. In order for OMR to break free from those restrictions, we propose to jointly recover the density map and the orthogonal matrices by requiring that they be mutually consistent. We regularize the resulting non-convex optimization problem by a denoised reference projection and a nonnegativity constraint. This is enabled by the new closed-form expressions for spatial autocorrelation features. Further, we design an easy-to-compute initial density map which effectively mitigates the non-convexity of the reconstruction problem. Experimental results show that the proposed OMR with spatial consensus is more robust and performs significantly better than the previous state-of-the-art OMR approach in the typical low-SNR scenario of 3D UVT.

64.Context-aware Self-supervised Learning for Medical Images Using Graph Neural Network ⬇️

Although self-supervised learning enables us to bootstrap the training by exploiting unlabeled data, the generic self-supervised methods for natural images do not sufficiently incorporate the context. For medical images, a desirable method should be sensitive enough to detect deviation from normal-appearing tissue of each anatomical region; here, anatomy is the context. We introduce a novel approach with two levels of self-supervised representation learning objectives: one on the regional anatomical level and another on the patient-level. We use graph neural networks to incorporate the relationship between different anatomical regions. The structure of the graph is informed by anatomical correspondences between each patient and an anatomical atlas. In addition, the graph representation has the advantage of handling any arbitrarily sized image in full resolution. Experiments on large-scale Computer Tomography (CT) datasets of lung images show that our approach compares favorably to baseline methods that do not account for the context. We use the learned embedding for staging lung tissue abnormalities related to COVID-19.

65.Virtual staining of defocused autofluorescence images of unlabeled tissue using deep neural networks ⬇️

Deep learning-based virtual staining was developed to introduce image contrast to label-free tissue sections, digitally matching the histological staining, which is time-consuming, labor-intensive, and destructive to tissue. Standard virtual staining requires high autofocusing precision during the whole slide imaging of label-free tissue, which consumes a significant portion of the total imaging time and can lead to tissue photodamage. Here, we introduce a fast virtual staining framework that can stain defocused autofluorescence images of unlabeled tissue, achieving equivalent performance to virtual staining of in-focus label-free images, also saving significant imaging time by lowering the microscope's autofocusing precision. This framework incorporates a virtual-autofocusing neural network to digitally refocus the defocused images and then transforms the refocused images into virtually stained images using a successive network. These cascaded networks form a collaborative inference scheme: the virtual staining model regularizes the virtual-autofocusing network through a style loss during the training. To demonstrate the efficacy of this framework, we trained and blindly tested these networks using human lung tissue. Using 4x fewer focus points with 2x lower focusing precision, we successfully transformed the coarsely-focused autofluorescence images into high-quality virtually stained H&E images, matching the standard virtual staining framework that used finely-focused autofluorescence input images. Without sacrificing the staining quality, this framework decreases the total image acquisition time needed for virtual staining of a label-free whole-slide image (WSI) by ~32%, together with a ~89% decrease in the autofocusing time, and has the potential to eliminate the laborious and costly histochemical staining process in pathology.

66.Humans Social Relationship Classification during Accompaniment ⬇️

This paper presents the design of deep learning architectures which allow to classify the social relationship existing between two people who are walking in a side-by-side formation into four possible categories --colleagues, couple, family or friendship. The models are developed using Neural Networks or Recurrent Neural Networks to achieve the classification and are trained and evaluated using a database of readings obtained from humans performing an accompaniment process in an urban environment. The best achieved model accomplishes a relatively good accuracy in the classification problem and its results enhance partially the outcomes from a previous study [1]. Furthermore, the model proposed shows its future potential to improve its efficiency and to be implemented in a real robot.

67.Physical Interaction and Manipulation of the Environment using Aerial Robots ⬇️

The physical interaction of aerial robots with their environment has countless potential applications and is an emerging area with many open challenges. Fully-actuated multirotors have been introduced to tackle some of these challenges. They provide complete control over position and orientation and eliminate the need for attaching a multi-DoF manipulation arm to the robot. However, there are many open problems before they can be used in real-world applications. Researchers have introduced some methods for physical interaction in limited settings. Their experiments primarily use prototype-level software without an efficient path to integration with real-world applications. We describe a new cost-effective solution for integrating these robots with the existing software and hardware flight systems for real-world applications and expand it to physical interaction applications. On the other hand, the existing control approaches for fully-actuated robots assume conservative limits for the thrusts and moments available to the robot. Using conservative assumptions for these already-inefficient robots makes their interactions even less optimal and may even result in many feasible physical interaction applications becoming infeasible. This work proposes a real-time method for estimating the complete set of instantaneously available forces and moments that robots can use to optimize their physical interaction performance. Finally, many real-world applications where aerial robots can improve the existing manual solutions deal with deformable objects. However, the perception and planning for their manipulation is still challenging. This research explores how aerial physical interaction can be extended to deformable objects. It provides a detection method suitable for manipulating deformable one-dimensional objects and introduces a new perspective on planning the manipulation of these objects.

68.Perfusion imaging in deep prostate cancer detection from mp-MRI: can we take advantage of it? ⬇️

To our knowledge, all deep computer-aided detection and diagnosis (CAD) systems for prostate cancer (PCa) detection consider bi-parametric magnetic resonance imaging (bp-MRI) only, including T2w and ADC sequences while excluding the 4D perfusion sequence,which is however part of standard clinical protocols for this diagnostic task. In this paper, we question strategies to integrate information from perfusion imaging in deep neural architectures. To do so, we evaluate several ways to encode the perfusion information in a U-Net like architecture, also considering early versus mid fusion strategies. We compare performance of multiparametric MRI (mp-MRI) models with the baseline bp-MRI model based on a private dataset of 219 mp-MRI exams. Perfusion maps derived from dynamic contrast enhanced MR exams are shown to positively impact segmentation and grading performance of PCa lesions, especially the 3D MR volume corresponding to the maximum slope of the wash-in curve as well as Tmax perfusion maps. The latter mp-MRI models indeed outperform the bp-MRI one whatever the fusion strategy, with Cohen's kappa score of 0.318$\pm$0.019 for the bp-MRI model and 0.378 $\pm$ 0.033 for the model including the maximum slope with a mid fusion strategy, also achieving competitive Cohen's kappa score compared to state of the art.