20220615.md

ArXiv cs.CV --Wed, 15 Jun 2022

1.RGB-Multispectral Matching: Dataset, Learning Methodology, Evaluation ⬇️

We address the problem of registering synchronized color (RGB) and multi-spectral (MS) images featuring very different resolution by solving stereo matching correspondences. Purposely, we introduce a novel RGB-MS dataset framing 13 different scenes in indoor environments and providing a total of 34 image pairs annotated with semi-dense, high-resolution ground-truth labels in the form of disparity maps. To tackle the task, we propose a deep learning architecture trained in a self-supervised manner by exploiting a further RGB camera, required only during training data acquisition. In this setup, we can conveniently learn cross-modal matching in the absence of ground-truth labels by distilling knowledge from an easier RGB-RGB matching task based on a collection of about 11K unlabeled image triplets. Experiments show that the proposed pipeline sets a good performance bar (1.16 pixels average registration error) for future research on this novel, challenging task.

2.ReCo: Retrieve and Co-segment for Zero-shot Transfer ⬇️

Semantic segmentation has a broad range of applications, but its real-world impact has been significantly limited by the prohibitive annotation costs necessary to enable deployment. Segmentation methods that forgo supervision can side-step these costs, but exhibit the inconvenient requirement to provide labelled examples from the target distribution to assign concept names to predictions. An alternative line of work in language-image pre-training has recently demonstrated the potential to produce models that can both assign names across large vocabularies of concepts and enable zero-shot transfer for classification, but do not demonstrate commensurate segmentation abilities. In this work, we strive to achieve a synthesis of these two approaches that combines their strengths. We leverage the retrieval abilities of one such language-image pre-trained model, CLIP, to dynamically curate training sets from unlabelled images for arbitrary collections of concept names, and leverage the robust correspondences offered by modern image representations to co-segment entities among the resulting collections. The synthetic segment collections are then employed to construct a segmentation model (without requiring pixel labels) whose knowledge of concepts is inherited from the scalable pre-training process of CLIP. We demonstrate that our approach, termed Retrieve and Co-segment (ReCo) performs favourably to unsupervised segmentation approaches while inheriting the convenience of nameable predictions and zero-shot transfer. We also demonstrate ReCo's ability to generate specialist segmenters for extremely rare objects.

3.AnimeSR: Learning Real-World Super-Resolution Models for Animation Videos ⬇️

This paper studies the problem of real-world video super-resolution (VSR) for animation videos, and reveals three key improvements for practical animation VSR. First, recent real-world super-resolution approaches typically rely on degradation simulation using basic operators without any learning capability, such as blur, noise, and compression. In this work, we propose to learn such basic operators from real low-quality animation videos, and incorporate the learned ones into the degradation generation pipeline. Such neural-network-based basic operators could help to better capture the distribution of real degradations. Second, a large-scale high-quality animation video dataset, AVC, is built to facilitate comprehensive training and evaluations for animation VSR. Third, we further investigate an efficient multi-scale network structure. It takes advantage of the efficiency of unidirectional recurrent networks and the effectiveness of sliding-window-based methods. Thanks to the above delicate designs, our method, AnimeSR, is capable of restoring real-world low-quality animation videos effectively and efficiently, achieving superior performance to previous state-of-the-art methods.

4.Accurate 3D Body Shape Regression using Metric and Semantic Attributes ⬇️

While methods that regress 3D human meshes from images have progressed rapidly, the estimated body shapes often do not capture the true human shape. This is problematic since, for many applications, accurate body shape is as important as pose. The key reason that body shape accuracy lags pose accuracy is the lack of data. While humans can label 2D joints, and these constrain 3D pose, it is not so easy to "label" 3D body shape. Since paired data with images and 3D body shape are rare, we exploit two sources of information: (1) we collect internet images of diverse "fashion" models together with a small set of anthropometric measurements; (2) we collect linguistic shape attributes for a wide range of 3D body meshes and the model images. Taken together, these datasets provide sufficient constraints to infer dense 3D shape. We exploit the anthropometric measurements and linguistic shape attributes in several novel ways to train a neural network, called SHAPY, that regresses 3D human pose and shape from an RGB image. We evaluate SHAPY on public benchmarks, but note that they either lack significant body shape variation, ground-truth shape, or clothing variation. Thus, we collect a new dataset for evaluating 3D human shape estimation, called HBW, containing photos of "Human Bodies in the Wild" for which we have ground-truth 3D body scans. On this new benchmark, SHAPY significantly outperforms state-of-the-art methods on the task of 3D body shape estimation. This is the first demonstration that 3D body shape regression from images can be trained from easy-to-obtain anthropometric measurements and linguistic shape attributes. Our model and data are available at: this http URL

5.Learning 3D Object Shape and Layout without 3D Supervision ⬇️

A 3D scene consists of a set of objects, each with a shape and a layout giving their position in space. Understanding 3D scenes from 2D images is an important goal, with applications in robotics and graphics. While there have been recent advances in predicting 3D shape and layout from a single image, most approaches rely on 3D ground truth for training which is expensive to collect at scale. We overcome these limitations and propose a method that learns to predict 3D shape and layout for objects without any ground truth shape or layout information: instead we rely on multi-view images with 2D supervision which can more easily be collected at scale. Through extensive experiments on 3D Warehouse, Hypersim, and ScanNet we demonstrate that our approach scales to large datasets of realistic images, and compares favorably to methods relying on 3D ground truth. On Hypersim and ScanNet where reliable 3D ground truth is not available, our approach outperforms supervised approaches trained on smaller and less diverse datasets.

6.Turning a Curse Into a Blessing: Enabling Clean-Data-Free Defenses by Model Inversion ⬇️

It is becoming increasingly common to utilize pre-trained models provided by third parties due to their convenience. At the same time, however, these models may be vulnerable to both poisoning and evasion attacks. We introduce an algorithmic framework that can mitigate potential security vulnerabilities in a pre-trained model when clean data from its training distribution is unavailable to the defender. The framework reverse-engineers samples from a given pre-trained model. The resulting synthetic samples can then be used as a substitute for clean data to perform various defenses. We consider two important attack scenarios -- backdoor attacks and evasion attacks -- to showcase the utility of synthesized samples. For both attacks, we show that when supplied with our synthetic data, the state-of-the-art defenses perform comparably or sometimes even better than the case when it's supplied with the same amount of clean data.

7.Consistent Video Instance Segmentation with Inter-Frame Recurrent Attention ⬇️

Video instance segmentation aims at predicting object segmentation masks for each frame, as well as associating the instances across multiple frames. Recent end-to-end video instance segmentation methods are capable of performing object segmentation and instance association together in a direct parallel sequence decoding/prediction framework. Although these methods generally predict higher quality object segmentation masks, they can fail to associate instances in challenging cases because they do not explicitly model the temporal instance consistency for adjacent frames. We propose a consistent end-to-end video instance segmentation framework with Inter-Frame Recurrent Attention to model both the temporal instance consistency for adjacent frames and the global temporal context. Our extensive experiments demonstrate that the Inter-Frame Recurrent Attention significantly improves temporal instance consistency while maintaining the quality of the object segmentation masks. Our model achieves state-of-the-art accuracy on both YouTubeVIS-2019 (62.1%) and YouTubeVIS-2021 (54.7%) datasets. In addition, quantitative and qualitative results show that the proposed methods predict more temporally consistent instance segmentation masks.

8.AuxMix: Semi-Supervised Learning with Unconstrained Unlabeled Data ⬇️

Semi-supervised learning (SSL) has seen great strides when labeled data is scarce but unlabeled data is abundant. Critically, most recent work assume that such unlabeled data is drawn from the same distribution as the labeled data. In this work, we show that state-of-the-art SSL algorithms suffer a degradation in performance in the presence of unlabeled auxiliary data that does not necessarily possess the same class distribution as the labeled set. We term this problem as Auxiliary-SSL and propose AuxMix, an algorithm that leverages self-supervised learning tasks to learn generic features in order to mask auxiliary data that are not semantically similar to the labeled set. We also propose to regularize learning by maximizing the predicted entropy for dissimilar auxiliary samples. We show an improvement of 5% over existing baselines on a ResNet-50 model when trained on CIFAR10 dataset with 4k labeled samples and all unlabeled data is drawn from the Tiny-ImageNet dataset. We report competitive results on several datasets and conduct ablation studies.

9.Monitoring Urban Forests from Auto-Generated Segmentation Maps ⬇️

We present and evaluate a weakly-supervised methodology to quantify the spatio-temporal distribution of urban forests based on remotely sensed data with close-to-zero human interaction. Successfully training machine learning models for semantic segmentation typically depends on the availability of high-quality labels. We evaluate the benefit of high-resolution, three-dimensional point cloud data (LiDAR) as source of noisy labels in order to train models for the localization of trees in orthophotos. As proof of concept we sense Hurricane Sandy's impact on urban forests in Coney Island, New York City (NYC) and reference it to less impacted urban space in Brooklyn, NYC.

10.Stand-Alone Inter-Frame Attention in Video Models ⬇️

Motion, as the uniqueness of a video, has been critical to the development of video understanding models. Modern deep learning models leverage motion by either executing spatio-temporal 3D convolutions, factorizing 3D convolutions into spatial and temporal convolutions separately, or computing self-attention along temporal dimension. The implicit assumption behind such successes is that the feature maps across consecutive frames can be nicely aggregated. Nevertheless, the assumption may not always hold especially for the regions with large deformation. In this paper, we present a new recipe of inter-frame attention block, namely Stand-alone Inter-Frame Attention (SIFA), that novelly delves into the deformation across frames to estimate local self-attention on each spatial location. Technically, SIFA remoulds the deformable design via re-scaling the offset predictions by the difference between two frames. Taking each spatial location in the current frame as the query, the locally deformable neighbors in the next frame are regarded as the keys/values. Then, SIFA measures the similarity between query and keys as stand-alone attention to weighted average the values for temporal aggregation. We further plug SIFA block into ConvNets and Vision Transformer, respectively, to devise SIFA-Net and SIFA-Transformer. Extensive experiments conducted on four video datasets demonstrate the superiority of SIFA-Net and SIFA-Transformer as stronger backbones. More remarkably, SIFA-Transformer achieves an accuracy of 83.1% on Kinetics-400 dataset. Source code is available at \url{this https URL}.

11.Comprehending and Ordering Semantics for Image Captioning ⬇️

Comprehending the rich semantics in an image and ordering them in linguistic order are essential to compose a visually-grounded and linguistically coherent description for image captioning. Modern techniques commonly capitalize on a pre-trained object detector/classifier to mine the semantics in an image, while leaving the inherent linguistic ordering of semantics under-exploited. In this paper, we propose a new recipe of Transformer-style structure, namely Comprehending and Ordering Semantics Networks (COS-Net), that novelly unifies an enriched semantic comprehending and a learnable semantic ordering processes into a single architecture. Technically, we initially utilize a cross-modal retrieval model to search the relevant sentences of each image, and all words in the searched sentences are taken as primary semantic cues. Next, a novel semantic comprehender is devised to filter out the irrelevant semantic words in primary semantic cues, and meanwhile infer the missing relevant semantic words visually grounded in the image. After that, we feed all the screened and enriched semantic words into a semantic ranker, which learns to allocate all semantic words in linguistic order as humans. Such sequence of ordered semantic words are further integrated with visual tokens of images to trigger sentence generation. Empirical evidences show that COS-Net clearly surpasses the state-of-the-art approaches on COCO and achieves to-date the best CIDEr score of 141.1% on Karpathy test split. Source code is available at \url{this https URL}.

12.A Multi-task Framework for Infrared Small Target Detection and Segmentation ⬇️

Due to the complicated background and noise of infrared images, infrared small target detection is one of the most difficult problems in the field of computer vision. In most existing studies, semantic segmentation methods are typically used to achieve better results. The centroid of each target is calculated from the segmentation map as the detection result. In contrast, we propose a novel end-to-end framework for infrared small target detection and segmentation in this paper. First, with the use of UNet as the backbone to maintain resolution and semantic information, our model can achieve a higher detection accuracy than other state-of-the-art methods by attaching a simple anchor-free head. Then, a pyramid pool module is used to further extract features and improve the precision of target segmentation. Next, we use semantic segmentation tasks that pay more attention to pixel-level features to assist in the training process of object detection, which increases the average precision and allows the model to detect some targets that were previously not detectable. Furthermore, we develop a multi-task framework for infrared small target detection and segmentation. Our multi-task learning model reduces complexity by nearly half and speeds up inference by nearly twice compared to the composite single-task model, while maintaining accuracy. The code and models are publicly available at this https URL.

13.Object Scene Representation Transformer ⬇️

A compositional understanding of the world in terms of objects and their geometry in 3D space is considered a cornerstone of human cognition. Facilitating the learning of such a representation in neural networks holds promise for substantially improving labeled data efficiency. As a key step in this direction, we make progress on the problem of learning 3D-consistent decompositions of complex scenes into individual objects in an unsupervised fashion. We introduce Object Scene Representation Transformer (OSRT), a 3D-centric model in which individual object representations naturally emerge through novel view synthesis. OSRT scales to significantly more complex scenes with larger diversity of objects and backgrounds than existing methods. At the same time, it is multiple orders of magnitude faster at compositional rendering thanks to its light field parametrization and the novel Slot Mixer decoder. We believe this work will not only accelerate future architecture exploration and scaling efforts, but it will also serve as a useful tool for both object-centric as well as neural scene representation learning communities.

14.Efficient Decoder-free Object Detection with Transformers ⬇️

Vision transformers (ViTs) are changing the landscape of object detection approaches. A natural usage of ViTs in detection is to replace the CNN-based backbone with a transformer-based backbone, which is straightforward and effective, with the price of bringing considerable computation burden for inference. More subtle usage is the DETR family, which eliminates the need for many hand-designed components in object detection but introduces a decoder demanding an extra-long time to converge. As a result, transformer-based object detection can not prevail in large-scale applications. To overcome these issues, we propose a novel decoder-free fully transformer-based (DFFT) object detector, achieving high efficiency in both training and inference stages, for the first time. We simplify objection detection into an encoder-only single-level anchor-based dense prediction problem by centering around two entry points: 1) Eliminate the training-inefficient decoder and leverage two strong encoders to preserve the accuracy of single-level feature map prediction; 2) Explore low-level semantic features for the detection task with limited computational resources. In particular, we design a novel lightweight detection-oriented transformer backbone that efficiently captures low-level features with rich semantics based on a well-conceived ablation study. Extensive experiments on the MS COCO benchmark demonstrate that DFFT_SMALL outperforms DETR by 2.5% AP with 28% computation cost reduction and more than $10$x fewer training epochs. Compared with the cutting-edge anchor-based detector RetinaNet, DFFT_SMALL obtains over 5.5% AP gain while cutting down 70% computation cost.

15.Asymmetric Dual-Decoder U-Net for Joint Rain and Haze Removal ⬇️

This work studies the joint rain and haze removal problem. In real-life scenarios, rain and haze, two often co-occurring common weather phenomena, can greatly degrade the clarity and quality of the scene images, leading to a performance drop in the visual applications, such as autonomous driving. However, jointly removing the rain and haze in scene images is ill-posed and challenging, where the existence of haze and rain and the change of atmosphere light, can both degrade the scene information. Current methods focus on the contamination removal part, thus ignoring the restoration of the scene information affected by the change of atmospheric light. We propose a novel deep neural network, named Asymmetric Dual-decoder U-Net (ADU-Net), to address the aforementioned challenge. The ADU-Net produces both the contamination residual and the scene residual to efficiently remove the rain and haze while preserving the fidelity of the scene information. Extensive experiments show our work outperforms the existing state-of-the-art methods by a considerable margin in both synthetic data and real-world data benchmarks, including RainCityscapes, BID Rain, and SPA-Data. For instance, we improve the state-of-the-art PSNR value by 2.26/4.57 on the RainCityscapes/SPA-Data, respectively.
Codes will be made available freely to the research community.

16.Peripheral Vision Transformer ⬇️

Human vision possesses a special type of visual processing systems called peripheral vision. Partitioning the entire visual field into multiple contour regions based on the distance to the center of our gaze, the peripheral vision provides us the ability to perceive various visual features at different regions. In this work, we take a biologically inspired approach and explore to model peripheral vision in deep neural networks for visual recognition. We propose to incorporate peripheral position encoding to the multi-head self-attention layers to let the network learn to partition the visual field into diverse peripheral regions given training data. We evaluate the proposed network, dubbed PerViT, on the large-scale ImageNet dataset and systematically investigate the inner workings of the model for machine perception, showing that the network learns to perceive visual data similarly to the way that human vision does. The state-of-the-art performance in image classification task across various model sizes demonstrates the efficacy of the proposed method.

17.Exploring Adversarial Attacks and Defenses in Vision Transformers trained with DINO ⬇️

This work conducts the first analysis on the robustness against adversarial attacks on self-supervised Vision Transformers trained using DINO. First, we evaluate whether features learned through self-supervision are more robust to adversarial attacks than those emerging from supervised learning. Then, we present properties arising for attacks in the latent space. Finally, we evaluate whether three well-known defense strategies can increase adversarial robustness in downstream tasks by only fine-tuning the classification head to provide robustness even in view of limited compute resources. These defense strategies are: Adversarial Training, Ensemble Adversarial Training and Ensemble of Specialized Networks.

18.Weakly-Supervised Crack Detection ⬇️

Pixel-level crack segmentation is widely studied due to its high impact on building and road inspections. Recent studies have made significant improvements in accuracy, but overlooked the annotation cost bottleneck. To resolve this issue, we reformulate the crack segmentation problem as a weakly-supervised problem, and propose a two-branched inference framework and an annotation refinement module that requires no additional data, in order to counteract the loss in annotation quality. Experimental results confirm the effectiveness of the proposed method in crack segmentation as well as other target domains.

19.Recurrent Transformer Variational Autoencoders for Multi-Action Motion Synthesis ⬇️

We consider the problem of synthesizing multi-action human motion sequences of arbitrary lengths. Existing approaches have mastered motion sequence generation in single-action scenarios, but fail to generalize to multi-action and arbitrary-length sequences. We fill this gap by proposing a novel efficient approach that leverages the expressiveness of Recurrent Transformers and generative richness of conditional Variational Autoencoders. The proposed iterative approach is able to generate smooth and realistic human motion sequences with an arbitrary number of actions and frames while doing so in linear space and time. We train and evaluate the proposed approach on PROX dataset which we augment with ground-truth action labels. Experimental evaluation shows significant improvements in FID score and semantic consistency metrics compared to the state-of-the-art.

20.Learning Dense Features for Point Cloud Registration Using Graph Attention Network ⬇️

Point cloud registration is a fundamental task in many applications such as localization, mapping, tracking, and reconstruction. The successful registration relies on extracting robust and discriminative geometric features. Existing learning-based methods require high computing capacity for processing a large number of raw points at the same time. Although these approaches achieve convincing results, they are difficult to apply in real-world situations due to high computational costs. In this paper, we introduce a framework that efficiently and economically extracts dense features using graph attention network for point cloud matching and registration (DFGAT). The detector of the DFGAT is responsible for finding highly reliable key points in large raw data sets. The descriptor of the DFGAT takes these key points combined with their neighbors to extract invariant density features in preparation for the matching. The graph attention network uses the attention mechanism that enriches the relationships between point clouds. Finally, we consider this as an optimal transport problem and use the Sinkhorn algorithm to find positive and negative matches. We perform thorough tests on the KITTI dataset and evaluate the effectiveness of this approach. The results show that this method with the efficiently compact keypoint selection and description can achieve the best performance matching metrics and reach highest success ratio of 99.88% registration in comparison with other state-of-the-art approaches.

21.Semi-signed neural fitting for surface reconstruction from unoriented point clouds ⬇️

Reconstructing 3D geometry from \emph{unoriented} point clouds can benefit many downstream tasks. Recent methods mostly adopt a neural shape representation with a neural network to represent a signed distance field and fit the point cloud with an unsigned supervision. However, we observe that using unsigned supervision may cause severe ambiguities and often leads to \emph{unexpected} failures such as generating undesired surfaces in free space when reconstructing complex structures and struggle with reconstructing accurate surfaces. To reconstruct a better signed distance field, we propose semi-signed neural fitting (SSN-Fitting), which consists of a semi-signed supervision and a loss-based region sampling strategy. Our key insight is that signed supervision is more informative and regions that are obviously outside the object can be easily determined. Meanwhile, a novel importance sampling is proposed to accelerate the optimization and better reconstruct the fine details. Specifically, we voxelize and partition the object space into \emph{sign-known} and \emph{sign-uncertain} regions, in which different supervisions are applied. Also, we adaptively adjust the sampling rate of each voxel according to the tracked reconstruction loss, so that the network can focus more on the complex under-fitting regions. We conduct extensive experiments to demonstrate that SSN-Fitting achieves state-of-the-art performance under different settings on multiple datasets, including clean, density-varying, and noisy data.

22.Interpretable Gait Recognition by Granger Causality ⬇️

Which joint interactions in the human gait cycle can be used as biometric characteristics? Most current methods on gait recognition suffer from the lack of interpretability. We propose an interpretable feature representation of gait sequences by the graphical Granger causal inference. Gait sequence of a person in the standardized motion capture format, constituting a set of 3D joint spatial trajectories, is envisaged as a causal system of joints interacting in time. We apply the graphical Granger model (GGM) to obtain the so-called Granger causal graph among joints as a discriminative and visually interpretable representation of a person's gait. We evaluate eleven distance functions in the GGM feature space by established classification and class-separability evaluation metrics. Our experiments indicate that, depending on the metric, the most appropriate distance functions for the GGM are the total norm distance and the Ky-Fan 1-norm distance. Experiments also show that the GGM is able to detect the most discriminative joint interactions and that it outperforms five related interpretable models in correct classification rate and in Davies-Bouldin index. The proposed GGM model can serve as a complementary tool for gait analysis in kinesiology or for gait recognition in video surveillance.

23.Visual Radial Basis Q-Network ⬇️

While reinforcement learning (RL) from raw images has been largely investigated in the last decade, existing approaches still suffer from a number of constraints. The high input dimension is often handled using either expert knowledge to extract handcrafted features or environment encoding through convolutional networks. Both solutions require numerous parameters to be optimized. In contrast, we propose a generic method to extract sparse features from raw images with few trainable parameters. We achieved this using a Radial Basis Function Network (RBFN) directly on raw image. We evaluate the performance of the proposed approach for visual extraction in Q-learning tasks in the Vizdoom environment. Then, we compare our results with two Deep Q-Network, one trained directly on images and another one trained on feature extracted by a pretrained auto-encoder. We show that the proposed approach provides similar or, in some cases, even better performances with fewer trainable parameters while being conceptually simpler.

24.ISLES 2022: A multi-center magnetic resonance imaging stroke lesion segmentation dataset ⬇️

Magnetic resonance imaging (MRI) is a central modality for stroke imaging. It is used upon patient admission to make treatment decisions such as selecting patients for intravenous thrombolysis or endovascular therapy. MRI is later used in the duration of hospital stay to predict outcome by visualizing infarct core size and location. Furthermore, it may be used to characterize stroke etiology, e.g. differentiation between (cardio)-embolic and non-embolic stroke. Computer based automated medical image processing is increasingly finding its way into clinical routine. Previous iterations of the Ischemic Stroke Lesion Segmentation (ISLES) challenge have aided in the generation of identifying benchmark methods for acute and sub-acute ischemic stroke lesion segmentation. Here we introduce an expert-annotated, multicenter MRI dataset for segmentation of acute to subacute stroke lesions. This dataset comprises 400 multi-vendor MRI cases with high variability in stroke lesion size, quantity and location. It is split into a training dataset of n=250 and a test dataset of n=150. All training data will be made publicly available. The test dataset will be used for model validation only and will not be released to the public. This dataset serves as the foundation of the ISLES 2022 challenge with the goal of finding algorithmic methods to enable the development and benchmarking of robust and accurate segmentation algorithms for ischemic stroke.

25.Online Easy Example Mining for Weakly-supervised Gland Segmentation from Histology Images ⬇️

Developing an AI-assisted gland segmentation method from histology images is critical for automatic cancer diagnosis and prognosis; however, the high cost of pixel-level annotations hinders its applications to broader diseases. Existing weakly-supervised semantic segmentation methods in computer vision achieve degenerative results for gland segmentation, since the characteristics and problems of glandular datasets are different from general object datasets. We observe that, unlike natural images, the key problem with histology images is the confusion of classes owning to morphological homogeneity and low color contrast among different tissues. To this end, we propose a novel method Online Easy Example Mining (OEEM) that encourages the network to focus on credible supervision signals rather than noisy signals, therefore mitigating the influence of inevitable false predictions in pseudo-masks. According to the characteristics of glandular datasets, we design a strong framework for gland segmentation. Our results exceed many fully-supervised methods and weakly-supervised methods for gland segmentation over 4.4% and 6.04% at mIoU, respectively. Code is available at this https URL.

26.Confidence Score for Source-Free Unsupervised Domain Adaptation ⬇️

Source-free unsupervised domain adaptation (SFUDA) aims to obtain high performance in the unlabeled target domain using the pre-trained source model, not the source data. Existing SFUDA methods assign the same importance to all target samples, which is vulnerable to incorrect pseudo-labels. To differentiate between sample importance, in this study, we propose a novel sample-wise confidence score, the Joint Model-Data Structure (JMDS) score for SFUDA. Unlike existing confidence scores that use only one of the source or target domain knowledge, the JMDS score uses both knowledge. We then propose a Confidence score Weighting Adaptation using the JMDS (CoWA-JMDS) framework for SFUDA. CoWA-JMDS consists of the JMDS scores as sample weights and weight Mixup that is our proposed variant of Mixup. Weight Mixup promotes the model make more use of the target domain knowledge. The experimental results show that the JMDS score outperforms the existing confidence scores. Moreover, CoWA-JMDS achieves state-of-the-art performance on various SFUDA scenarios: closed, open, and partial-set scenarios.

27.RF-Next: Efficient Receptive Field Search for Convolutional Neural Networks ⬇️

Temporal/spatial receptive fields of models play an important role in sequential/spatial tasks. Large receptive fields facilitate long-term relations, while small receptive fields help to capture the local details. Existing methods construct models with hand-designed receptive fields in layers. Can we effectively search for receptive field combinations to replace hand-designed patterns? To answer this question, we propose to find better receptive field combinations through a global-to-local search scheme. Our search scheme exploits both global search to find the coarse combinations and local search to get the refined receptive field combinations further. The global search finds possible coarse combinations other than human-designed patterns. On top of the global search, we propose an expectation-guided iterative local search scheme to refine combinations effectively. Our RF-Next models, plugging receptive field search to various models, boost the performance on many tasks, e.g., temporal action segmentation, object detection, instance segmentation, and speech synthesis. The source code is publicly available on this http URL.

28.Slimmable Domain Adaptation ⬇️

Vanilla unsupervised domain adaptation methods tend to optimize the model with fixed neural architecture, which is not very practical in real-world scenarios since the target data is usually processed by different resource-limited devices. It is therefore of great necessity to facilitate architecture adaptation across various devices. In this paper, we introduce a simple framework, Slimmable Domain Adaptation, to improve cross-domain generalization with a weight-sharing model bank, from which models of different capacities can be sampled to accommodate different accuracy-efficiency trade-offs. The main challenge in this framework lies in simultaneously boosting the adaptation performance of numerous models in the model bank. To tackle this problem, we develop a Stochastic EnsEmble Distillation method to fully exploit the complementary knowledge in the model bank for inter-model interaction. Nevertheless, considering the optimization conflict between inter-model interaction and intra-model adaptation, we augment the existing bi-classifier domain confusion architecture into an Optimization-Separated Tri-Classifier counterpart. After optimizing the model bank, architecture adaptation is leveraged via our proposed Unsupervised Performance Evaluation Metric. Under various resource constraints, our framework surpasses other competing approaches by a very large margin on multiple benchmarks. It is also worth emphasizing that our framework can preserve the performance improvement against the source-only model even when the computing complexity is reduced to $1/64$. Code will be available at this https URL.

29.TransVG++: End-to-End Visual Grounding with Language Conditioned Vision Transformer ⬇️

In this work, we explore neat yet effective Transformer-based frameworks for visual grounding. The previous methods generally address the core problem of visual grounding, i.e., multi-modal fusion and reasoning, with manually-designed mechanisms. Such heuristic designs are not only complicated but also make models easily overfit specific data distributions. To avoid this, we first propose TransVG, which establishes multi-modal correspondences by Transformers and localizes referred regions by directly regressing box coordinates. We empirically show that complicated fusion modules can be replaced by a simple stack of Transformer encoder layers with higher performance. However, the core fusion Transformer in TransVG is stand-alone against uni-modal encoders, and thus should be trained from scratch on limited visual grounding data, which makes it hard to be optimized and leads to sub-optimal performance. To this end, we further introduce TransVG++ to make two-fold improvements. For one thing, we upgrade our framework to a purely Transformer-based one by leveraging Vision Transformer (ViT) for vision feature encoding. For another, we devise Language Conditioned Vision Transformer that removes external fusion modules and reuses the uni-modal ViT for vision-language fusion at the intermediate layers. We conduct extensive experiments on five prevalent datasets, and report a series of state-of-the-art records.

30.Label Matching Semi-Supervised Object Detection ⬇️

Semi-supervised object detection has made significant progress with the development of mean teacher driven self-training. Despite the promising results, the label mismatch problem is not yet fully explored in the previous works, leading to severe confirmation bias during self-training. In this paper, we delve into this problem and propose a simple yet effective LabelMatch framework from two different yet complementary perspectives, i.e., distribution-level and instance-level. For the former one, it is reasonable to approximate the class distribution of the unlabeled data from that of the labeled data according to Monte Carlo Sampling. Guided by this weakly supervision cue, we introduce a re-distribution mean teacher, which leverages adaptive label-distribution-aware confidence thresholds to generate unbiased pseudo labels to drive student learning. For the latter one, there exists an overlooked label assignment ambiguity problem across teacher-student models. To remedy this issue, we present a novel label assignment mechanism for self-training framework, namely proposal self-assignment, which injects the proposals from student into teacher and generates accurate pseudo labels to match each proposal in the student model accordingly. Experiments on both MS-COCO and PASCAL-VOC datasets demonstrate the considerable superiority of our proposed framework to other state-of-the-arts. Code will be available at this https URL.

31.Plug-and-Play Pseudo Label Correction Network for Unsupervised Person Re-identification ⬇️

Clustering-based methods, which alternate between the generation of pseudo labels and the optimization of the feature extraction network, play a dominant role in both unsupervised learning (USL) and unsupervised domain adaptive (UDA) person re-identification (Re-ID). To alleviate the adverse effect of noisy pseudo labels, the existing methods either abandon unreliable labels or refine the pseudo labels via mutual learning or label propagation. However, a great many erroneous labels are still accumulated because these methods mostly adopt traditional unsupervised clustering algorithms which rely on certain assumptions on data distribution and fail to capture the distribution of complex real-world data. In this paper, we propose the plug-and-play graph-based pseudo label correction network (GLC) to refine the pseudo labels in the manner of supervised clustering. GLC is trained to perceive the varying data distribution at each epoch of the self-training with the supervision of initial pseudo labels generated by any clustering method. It can learn to rectify the initial noisy labels by means of the relationship constraints between samples on the k Nearest Neighbor (kNN) graph and early-stop training strategy. Specifically, GLC learns to aggregate node features from neighbors and predict whether the nodes should be linked on the graph. Besides, GLC is optimized with 'early stop' before the noisy labels are severely memorized to prevent overfitting to noisy pseudo labels. Consequently, GLC improves the quality of pseudo labels though the supervision signals contain some noise, leading to better Re-ID performance. Extensive experiments in USL and UDA person Re-ID on Market-1501 and MSMT17 show that our method is widely compatible with various clustering-based methods and promotes the state-of-the-art performance consistently.

32.A Survey of Automated Data Augmentation Algorithms for Deep Learning-based Image Classication Tasks ⬇️

In recent years, one of the most popular techniques in the computer vision community has been the deep learning technique. As a data-driven technique, deep model requires enormous amounts of accurately labelled training data, which is often inaccessible in many real-world applications. A data-space solution is Data Augmentation (DA), that can artificially generate new images out of original samples. Image augmentation strategies can vary by dataset, as different data types might require different augmentations to facilitate model training. However, the design of DA policies has been largely decided by the human experts with domain knowledge, which is considered to be highly subjective and error-prone. To mitigate such problem, a novel direction is to automatically learn the image augmentation policies from the given dataset using Automated Data Augmentation (AutoDA) techniques. The goal of AutoDA models is to find the optimal DA policies that can maximize the model performance gains. This survey discusses the underlying reasons of the emergence of AutoDA technology from the perspective of image classification. We identify three key components of a standard AutoDA model: a search space, a search algorithm and an evaluation function. Based on their architecture, we provide a systematic taxonomy of existing image AutoDA approaches. This paper presents the major works in AutoDA field, discussing their pros and cons, and proposing several potential directions for future improvements.

33.3D scene reconstruction from monocular spherical video with motion parallax ⬇️

In this paper, we describe a method to capture nearly entirely spherical (360 degree) depth information using two adjacent frames from a single spherical video with motion parallax. After illustrating a spherical depth information retrieval using two spherical cameras, we demonstrate monocular spherical stereo by using stabilized first-person video footage. Experiments demonstrated that the depth information was retrieved on up to 97% of the entire sphere in solid angle. At a speed of 30 km/h, we were able to estimate the depth of an object located over 30 m from the camera. We also reconstructed the 3D structures (point cloud) using the obtained depth data and confirmed the structures can be clearly observed. We can apply this method to 3D structure retrieval of surrounding environments such as 1) previsualization, location hunting/planning of a film, 2) real scene/computer graphics synthesis and 3) motion capture. Thanks to its simplicity, this method can be applied to various videos. As there is no pre-condition other than to be a 360 video with motion parallax, we can use any 360 videos including those on the Internet to reconstruct the surrounding environments. The cameras can be lightweight enough to be mounted on a drone. We also demonstrated such applications.

34.Estimating Pose from Pressure Data for Smart Beds with Deep Image-based Pose Estimators ⬇️

In-bed pose estimation has shown value in fields such as hospital patient monitoring, sleep studies, and smart homes. In this paper, we explore different strategies for detecting body pose from highly ambiguous pressure data, with the aid of pre-existing pose estimators. We examine the performance of pre-trained pose estimators by using them either directly or by re-training them on two pressure datasets. We also explore other strategies utilizing a learnable pre-processing domain adaptation step, which transforms the vague pressure maps to a representation closer to the expected input space of common purpose pose estimation modules. Accordingly, we used a fully convolutional network with multiple scales to provide the pose-specific characteristics of the pressure maps to the pre-trained pose estimation module. Our complete analysis of different approaches shows that the combination of learnable pre-processing module along with re-training pre-existing image-based pose estimators on the pressure data is able to overcome issues such as highly vague pressure points to achieve very high pose estimation accuracy.

35.Generalizable Method for Face Anti-Spoofing with Semi-Supervised Learning ⬇️

Face anti-spoofing has drawn a lot of attention due to the high security requirements in biometric authentication systems. Bringing face biometric to commercial hardware became mostly dependent on developing reliable methods for detecting fake login sessions without specialized sensors. Current CNN-based method perform well on the domains they were trained for, but often show poor generalization on previously unseen datasets. In this paper we describe a method for utilizing unsupervised pretraining for improving performance across multiple datasets without any adaptation, introduce the Entry Antispoofing Dataset for supervised fine-tuning, and propose a multi-class auxiliary classification layer for augmenting the binary classification task of detecting spoofing attempts with explicit interpretable signals. We demonstrate the efficiency of our model by achieving state-of-the-art results on cross-dataset testing on MSU-MFSD, Replay-Attack, and OULU-NPU datasets.

36.Spiking Neural Networks for Frame-based and Event-based Single Object Localization ⬇️

Spiking neural networks have shown much promise as an energy-efficient alternative to artificial neural networks. However, understanding the impacts of sensor noises and input encodings on the network activity and performance remains difficult with common neuromorphic vision baselines like classification. Therefore, we propose a spiking neural network approach for single object localization trained using surrogate gradient descent, for frame- and event-based sensors. We compare our method with similar artificial neural networks and show that our model has competitive/better performance in accuracy, robustness against various corruptions, and has lower energy consumption. Moreover, we study the impact of neural coding schemes for static images in accuracy, robustness, and energy efficiency. Our observations differ importantly from previous studies on bio-plausible learning rules, which helps in the design of surrogate gradient trained architectures, and offers insight to design priorities in future neuromorphic technologies in terms of noise characteristics and data encoding methods.

37.Learning Task-Independent Game State Representations from Unlabeled Images ⬇️

Self-supervised learning (SSL) techniques have been widely used to learn compact and informative representations from high-dimensional complex data. In many computer vision tasks, such as image classification, such methods achieve state-of-the-art results that surpass supervised learning approaches. In this paper, we investigate whether SSL methods can be leveraged for the task of learning accurate state representations of games, and if so, to what extent. For this purpose, we collect game footage frames and corresponding sequences of games' internal state from three different 3D games: VizDoom, the CARLA racing simulator and the Google Research Football Environment. We train an image encoder with three widely used SSL algorithms using solely the raw frames, and then attempt to recover the internal state variables from the learned representations. Our results across all three games showcase significantly higher correlation between SSL representations and the game's internal state compared to pre-trained baseline models such as ImageNet. Such findings suggest that SSL-based visual encoders can yield general -- not tailored to a specific task -- yet informative game representations solely from game pixel information. Such representations can, in turn, form the basis for boosting the performance of downstream learning tasks in games, including gameplaying, content generation and player modeling.

38.Multimodal Learning with Transformers: A Survey ⬇️

Transformer is a promising neural network learner, and has achieved great success in various machine learning tasks. Thanks to the recent prevalence of multimodal applications and big data, Transformer-based multimodal learning has become a hot topic in AI research. This paper presents a comprehensive survey of Transformer techniques oriented at multimodal data. The main contents of this survey include: (1) a background of multimodal learning, Transformer ecosystem, and the multimodal big data era, (2) a theoretical review of Vanilla Transformer, Vision Transformer, and multimodal Transformers, from a geometrically topological perspective, (3) a review of multimodal Transformer applications, via two important paradigms, i.e., for multimodal pretraining and for specific multimodal tasks, (4) a summary of the common challenges and designs shared by the multimodal Transformer models and applications, and (5) a discussion of open problems and potential research directions for the community.

39.The Modality Focusing Hypothesis: On the Blink of Multimodal Knowledge Distillation ⬇️

Multimodal knowledge distillation (KD) extends traditional knowledge distillation to the area of multimodal learning. One common practice is to adopt a well-performed multimodal network as the teacher in the hope that it can transfer its full knowledge to a unimodal student for performance improvement. In this paper, we investigate the efficacy of multimodal KD. We begin by providing two failure cases of it and demonstrate that KD is not a universal cure in multimodal knowledge transfer. We present the modality Venn diagram to understand modality relationships and the modality focusing hypothesis revealing the decisive factor in the efficacy of multimodal KD. Experimental results on 6 multimodal datasets help justify our hypothesis, diagnose failure cases, and point directions to improve distillation performance.

40.On Image Segmentation With Noisy Labels: Characterization and Volume Properties of the Optimal Solutions to Accuracy and Dice ⬇️

We study two of the most popular performance metrics in medical image segmentation, Accuracy and Dice, when the target labels are noisy. For both metrics, several statements related to characterization and volume properties of the set of optimal segmentations are proved, and associated experiments are provided. Our main insights are: (i) the volume of the solutions to both metrics may deviate significantly from the expected volume of the target, (ii) the volume of a solution to Accuracy is always less than or equal to the volume of a solution to Dice and (iii) the optimal solutions to both of these metrics coincide when the set of feasible segmentations is constrained to the set of segmentations with the volume equal to the expected volume of the target.

41.RigNeRF: Fully Controllable Neural 3D Portraits ⬇️

Volumetric neural rendering methods, such as neural radiance fields (NeRFs), have enabled photo-realistic novel view synthesis. However, in their standard form, NeRFs do not support the editing of objects, such as a human head, within a scene. In this work, we propose RigNeRF, a system that goes beyond just novel view synthesis and enables full control of head pose and facial expressions learned from a single portrait video. We model changes in head pose and facial expressions using a deformation field that is guided by a 3D morphable face model (3DMM). The 3DMM effectively acts as a prior for RigNeRF that learns to predict only residuals to the 3DMM deformations and allows us to render novel (rigid) poses and (non-rigid) expressions that were not present in the input sequence. Using only a smartphone-captured short video of a subject for training, we demonstrate the effectiveness of our method on free view synthesis of a portrait scene with explicit head pose and expression controls. The project page can be found here: this http URL

42.Revisiting the Shape-Bias of Deep Learning for Dermoscopic Skin Lesion Classification ⬇️

It is generally believed that the human visual system is biased towards the recognition of shapes rather than textures. This assumption has led to a growing body of work aiming to align deep models' decision-making processes with the fundamental properties of human vision. The reliance on shape features is primarily expected to improve the robustness of these models under covariate shift. In this paper, we revisit the significance of shape-biases for the classification of skin lesion images. Our analysis shows that different skin lesion datasets exhibit varying biases towards individual image features. Interestingly, despite deep feature extractors being inclined towards learning entangled features for skin lesion classification, individual features can still be decoded from this entangled representation. This indicates that these features are still represented in the learnt embedding spaces of the models, but not used for classification. In addition, the spectral analysis of different datasets shows that in contrast to common visual recognition, dermoscopic skin lesion classification, by nature, is reliant on complex feature combinations beyond shape-bias. As a natural consequence, shifting away from the prevalent desire of shape-biasing models can even improve skin lesion classifiers in some cases.

43.Self-Supervised Representation Learning With MUlti-Segmental Informational Coding (MUSIC) ⬇️

Self-supervised representation learning maps high-dimensional data into a meaningful embedding space, where samples of similar semantic contents are close to each other. Most of the recent representation learning methods maximize cosine similarity or minimize the distance between the embedding features of different views from the same sample usually on the $l2$ normalized unit hypersphere. To prevent the trivial solutions that all samples have the same embedding feature, various techniques have been developed, such as contrastive learning, stop gradient, variance and covariance regularization, etc. In this study, we propose MUlti-Segmental Informational Coding (MUSIC) for self-supervised representation learning. MUSIC divides the embedding feature into multiple segments that discriminatively partition samples into different semantic clusters and different segments focus on different partition principles. Information theory measurements are directly used to optimize MUSIC and theoretically guarantee trivial solutions are avoided. MUSIC does not depend on commonly used techniques, such as memory bank or large batches, asymmetry networks, gradient stopping, momentum weight updating, etc, making the training framework flexible. Our experiments demonstrate that MUSIC achieves better results than most related Barlow Twins and VICReg methods on ImageNet classification with linear probing, and requires neither deep projectors nor large feature dimensions. Code will be made available.

44.ICP Algorithm: Theory, Practice And Its SLAM-oriented Taxonomy ⬇️

The Iterative Closest Point (ICP) algorithm is one of the most important algorithms for geometric alignment of three-dimensional surface registration, which is frequently used in computer vision tasks, including the Simultaneous Localization And Mapping (SLAM) tasks. In this paper, we illustrate the theoretical principles of the ICP algorithm, how it can be used in surface registration tasks, and the traditional taxonomy of the variants of the ICP algorithm. As SLAM is becoming a popular topic, we also introduce a SLAM-oriented taxonomy of the ICP algorithm, based on the characteristics of each type of SLAM task, including whether the SLAM task is online or not and whether the landmarks are present as features in the SLAM task. We make a synthesis of each type of SLAM task by comparing several up-to-date research papers and analyzing their implementation details.

45.A Training Method For VideoPose3D With Ideology of Action Recognition ⬇️

Action recognition and pose estimation from videos are closely related to understand human motions, but more literature focuses on how to solve pose estimation tasks alone from action recognition. This research shows a faster and more flexible training method for VideoPose3D which is based on action recognition. This model is fed with the same type of action as the type that will be estimated, and different types of actions can be trained separately. Evidence has shown that, for common pose-estimation tasks, this model requires a relatively small amount of data to carry out similar results with the original research, and for action-oriented tasks, it outperforms the original research by 4.5% with a limited receptive field size and training epoch on Velocity Error of MPJPE. This model can handle both action-oriented and common pose-estimation problems.

46.A Multi-purpose Real Haze Benchmark with Quantifiable Haze Levels and Ground Truth ⬇️

Imagery collected from outdoor visual environments is often degraded due to the presence of dense smoke or haze. A key challenge for research in scene understanding in these degraded visual environments (DVE) is the lack of representative benchmark datasets. These datasets are required to evaluate state-of-the-art object recognition and other computer vision algorithms in degraded settings. In this paper, we address some of these limitations by introducing the first paired real image benchmark dataset with hazy and haze-free images, and in-situ haze density measurements. This dataset was produced in a controlled environment with professional smoke generating machines that covered the entire scene, and consists of images captured from the perspective of both an unmanned aerial vehicle (UAV) and an unmanned ground vehicle (UGV). We also evaluate a set of representative state-of-the-art dehazing approaches as well as object detectors on the dataset. The full dataset presented in this paper, including the ground truth object classification bounding boxes and haze density measurements, is provided for the community to evaluate their algorithms at: https://a2i2-archangel.vision. A subset of this dataset has been used for the Object Detection in Haze Track of CVPR UG2 2022 challenge.

47.GraphMLP: A Graph MLP-Like Architecture for 3D Human Pose Estimation ⬇️

Modern multi-layer perceptron (MLP) models have shown competitive results in learning visual representations without self-attention. However, existing MLP models are not good at capturing local details and lack prior knowledge of human configurations, which limits their modeling power for skeletal representation learning. To address these issues, we propose a simple yet effective graph-reinforced MLP-Like architecture, named GraphMLP, that combines MLPs and graph convolutional networks (GCNs) in a global-local-graphical unified architecture for 3D human pose estimation. GraphMLP incorporates the graph structure of human bodies into an MLP model to meet the domain-specific demand while also allowing for both local and global spatial interactions. Extensive experiments show that the proposed GraphMLP achieves state-of-the-art performance on two datasets, i.e., Human3.6M and MPI-INF-3DHP. Our source code and pretrained models will be publicly available.

48.Compositional Mixture Representations for Vision and Text ⬇️

Learning a common representation space between vision and language allows deep networks to relate objects in the image to the corresponding semantic meaning. We present a model that learns a shared Gaussian mixture representation imposing the compositionality of the text onto the visual domain without having explicit location supervision. By combining the spatial transformer with a representation learning approach we learn to split images into separately encoded patches to associate visual and textual representations in an interpretable manner. On variations of MNIST and CIFAR10, our model is able to perform weakly supervised object detection and demonstrates its ability to extrapolate to unseen combination of objects.

49.ProcTHOR: Large-Scale Embodied AI Using Procedural Generation ⬇️

Massive datasets and high-capacity models have driven many recent advancements in computer vision and natural language understanding. This work presents a platform to enable similar success stories in Embodied AI. We propose ProcTHOR, a framework for procedural generation of Embodied AI environments. ProcTHOR enables us to sample arbitrarily large datasets of diverse, interactive, customizable, and performant virtual environments to train and evaluate embodied agents across navigation, interaction, and manipulation tasks. We demonstrate the power and potential of ProcTHOR via a sample of 10,000 generated houses and a simple neural model. Models trained using only RGB images on ProcTHOR, with no explicit mapping and no human task supervision produce state-of-the-art results across 6 embodied AI benchmarks for navigation, rearrangement, and arm manipulation, including the presently running Habitat 2022, AI2-THOR Rearrangement 2022, and RoboTHOR challenges. We also demonstrate strong 0-shot results on these benchmarks, via pre-training on ProcTHOR with no fine-tuning on the downstream benchmark, often beating previous state-of-the-art systems that access the downstream training data.

50.K-Space Transformer for Fast MRI Reconstruction with Implicit Representation ⬇️

This paper considers the problem of fast MRI reconstruction. We propose a novel Transformer-based framework for directly processing the sparsely sampled signals in k-space, going beyond the limitation of regular grids as ConvNets do. We adopt an implicit representation of spectrogram, treating spatial coordinates as inputs, and dynamically query the partially observed measurements to complete the spectrogram, i.e. learning the inductive bias in k-space. To strive a balance between computational cost and reconstruction quality, we build an hierarchical structure with low-resolution and high-resolution decoders respectively. To validate the necessity of our proposed modules, we have conducted extensive experiments on two public datasets, and demonstrate superior or comparable performance over state-of-the-art approaches.

51.Evaluating histopathology transfer learning with ChampKit ⬇️

Histopathology remains the gold standard for diagnosis of various cancers. Recent advances in computer vision, specifically deep learning, have facilitated the analysis of histopathology images for various tasks, including immune cell detection and microsatellite instability classification. The state-of-the-art for each task often employs base architectures that have been pretrained for image classification on ImageNet. The standard approach to develop classifiers in histopathology tends to focus narrowly on optimizing models for a single task, not considering the aspects of modeling innovations that improve generalization across tasks. Here we present ChampKit (Comprehensive Histopathology Assessment of Model Predictions toolKit): an extensible, fully reproducible benchmarking toolkit that consists of a broad collection of patch-level image classification tasks across different cancers. ChampKit enables a way to systematically document the performance impact of proposed improvements in models and methodology. ChampKit source code and data are freely accessible at this https URL .

52.When adversarial attacks become interpretable counterfactual explanations ⬇️

We argue that, when learning a 1-Lipschitz neural network with the dual loss of an optimal transportation problem, the gradient of the model is both the direction of the transportation plan and the direction to the closest adversarial attack. Traveling along the gradient to the decision boundary is no more an adversarial attack but becomes a counterfactual explanation, explicitly transporting from one class to the other. Through extensive experiments on XAI metrics, we find that the simple saliency map method, applied on such networks, becomes a reliable explanation, and outperforms the state-of-the-art explanation approaches on unconstrained models. The proposed networks were already known to be certifiably robust, and we prove that they are also explainable with a fast and simple method.

53.Learning towards Synchronous Network Memorizability and Generalizability for Continual Segmentation across Multiple Sites ⬇️

In clinical practice, a segmentation network is often required to continually learn on a sequential data stream from multiple sites rather than a consolidated set, due to the storage cost and privacy restriction. However, during the continual learning process, existing methods are usually restricted in either network memorizability on previous sites or generalizability on unseen sites. This paper aims to tackle the challenging problem of Synchronous Memorizability and Generalizability (SMG) and to simultaneously improve performance on both previous and unseen sites, with a novel proposed SMG-learning framework. First, we propose a Synchronous Gradient Alignment (SGA) objective, which \emph{not only} promotes the network memorizability by enforcing coordinated optimization for a small exemplar set from previous sites (called replay buffer), \emph{but also} enhances the generalizability by facilitating site-invariance under simulated domain shift. Second, to simplify the optimization of SGA objective, we design a Dual-Meta algorithm that approximates the SGA objective as dual meta-objectives for optimization without expensive computation overhead. Third, for efficient rehearsal, we configure the replay buffer comprehensively considering additional inter-site diversity to reduce redundancy. Experiments on prostate MRI data sequentially acquired from six institutes demonstrate that our method can simultaneously achieve higher memorizability and generalizability over state-of-the-art methods. Code is available at this https URL.

54.Adversarial Vulnerability of Randomized Ensembles ⬇️

Despite the tremendous success of deep neural networks across various tasks, their vulnerability to imperceptible adversarial perturbations has hindered their deployment in the real world. Recently, works on randomized ensembles have empirically demonstrated significant improvements in adversarial robustness over standard adversarially trained (AT) models with minimal computational overhead, making them a promising solution for safety-critical resource-constrained applications. However, this impressive performance raises the question: Are these robustness gains provided by randomized ensembles real? In this work we address this question both theoretically and empirically. We first establish theoretically that commonly employed robustness evaluation methods such as adaptive PGD provide a false sense of security in this setting. Subsequently, we propose a theoretically-sound and efficient adversarial attack algorithm (ARC) capable of compromising random ensembles even in cases where adaptive PGD fails to do so. We conduct comprehensive experiments across a variety of network architectures, training schemes, datasets, and norms to support our claims, and empirically establish that randomized ensembles are in fact more vulnerable to $\ell_p$-bounded adversarial perturbations than even standard AT models. Our code can be found at this https URL.

55.Automated Precision Localization of Peripherally Inserted Central Catheter Tip through Model-Agnostic Multi-Stage Networks ⬇️

Peripherally inserted central catheters (PICCs) have been widely used as one of the representative central venous lines (CVCs) due to their long-term intravascular access with low infectivity. However, PICCs have a fatal drawback of a high frequency of tip mispositions, increasing the risk of puncture, embolism, and complications such as cardiac arrhythmias. To automatically and precisely detect it, various attempts have been made by using the latest deep learning (DL) technologies. However, even with these approaches, it is still practically difficult to determine the tip location because the multiple fragments phenomenon (MFP) occurs in the process of predicting and extracting the PICC line required before predicting the tip. This study aimed to develop a system generally applied to existing models and to restore the PICC line more exactly by removing the MFs of the model output, thereby precisely localizing the actual tip position for detecting its disposition. To achieve this, we proposed a multi-stage DL-based framework post-processing the PICC line extraction result of the existing technology. The performance was compared by each root mean squared error (RMSE) and MFP incidence rate according to whether or not MFCN is applied to five conventional models. In internal validation, when MFCN was applied to the existing single model, MFP was improved by an average of 45%. The RMSE was improved by over 63% from an average of 26.85mm (17.16 to 35.80mm) to 9.72mm (9.37 to 10.98mm). In external validation, when MFCN was applied, the MFP incidence rate decreased by an average of 32% and the RMSE decreased by an average of 65%. Therefore, by applying the proposed MFCN, we observed the significant/consistent detection performance improvement of PICC tip location compared to the existing model.

56.Automated SSIM Regression for Detection and Quantification of Motion Artefacts in Brain MR Images ⬇️

Motion artefacts in magnetic resonance brain images are a crucial issue. The assessment of MR image quality is fundamental before proceeding with the clinical diagnosis. If the motion artefacts alter a correct delineation of structure and substructures of the brain, lesions, tumours and so on, the patients need to be re-scanned. Otherwise, neuro-radiologists could report an inaccurate or incorrect diagnosis. The first step right after scanning a patient is the "\textit{image quality assessment}" in order to decide if the acquired images are diagnostically acceptable. An automated image quality assessment based on the structural similarity index (SSIM) regression through a residual neural network has been proposed here, with the possibility to perform also the classification in different groups - by subdividing with SSIM ranges. This method predicts SSIM values of an input image in the absence of a reference ground truth image. The networks were able to detect motion artefacts, and the best performance for the regression and classification task has always been achieved with ResNet-18 with contrast augmentation. Mean and standard deviation of residuals' distribution were $\mu=-0.0009$ and $\sigma=0.0139$, respectively. Whilst for the classification task in 3, 5 and 10 classes, the best accuracies were 97, 95 and 89%, respectively. The obtained results show that the proposed method could be a tool in supporting neuro-radiologists and radiographers in evaluating the image quality before the diagnosis.

57.CNN-based Classification Framework for Tissues of Lung with Additional Information ⬇️

Interstitial lung diseases are a large group of heterogeneous diseases characterized by different degrees of alveolitis and pulmonary fibrosis. Accurately diagnosing these diseases has significant guiding value for formulating treatment plans. Although previous work has produced impressive results in classifying interstitial lung diseases, there is still room for improving the accuracy of these techniques, mainly to enhance automated decision-making. In order to improve the classification precision, our study proposes a convolutional neural networks-based framework with additional information. Firstly, ILD images are added with their medical information by re-scaling the original image in Hounsfield Units. Secondly, a modified CNN model is used to produce a vector of classification probability for each tissue. Thirdly, location information of the input image, consisting of the occurrence frequencies of different diseases in the CT scans on certain locations, is used to calculate a location weight vector. Finally, the Hadamard product between two vectors is used to produce a decision vector for the prediction. Compared to the state-of-the-art methods, the results using a publicly available ILD database show the potential of predicting these using different additional information.

58.Quantitative Imaging Principles Improves Medical Image Learning ⬇️

Fundamental differences between natural and medical images have recently favored the use of self-supervised learning (SSL) over ImageNet transfer learning for medical image applications. Differences between image types are primarily due to the imaging modality and medical images utilize a wide range of physics based techniques while natural images are captured using only visible light. While many have demonstrated that SSL on medical images has resulted in better downstream task performance, our work suggests that more performance can be gained. The scientific principles which are used to acquire medical images are not often considered when constructing learning problems. For this reason, we propose incorporating quantitative imaging principles during generative SSL to improve image quality and quantitative biological accuracy. We show that this training schema results in better starting states for downstream supervised training on limited data. Our model also generates images that validate on clinical quantitative analysis software.

59.Learning Best Combination for Efficient N:M Sparsity ⬇️

By forcing at most N out of M consecutive weights to be non-zero, the recent N:M network sparsity has received increasing attention for its two attractive advantages: 1) Promising performance at a high sparsity. 2) Significant speedups on NVIDIA A100 GPUs. Recent studies require an expensive pre-training phase or a heavy dense-gradient computation. In this paper, we show that the N:M learning can be naturally characterized as a combinatorial problem which searches for the best combination candidate within a finite collection. Motivated by this characteristic, we solve N:M sparsity in an efficient divide-and-conquer manner. First, we divide the weight vector into $C_{\text{M}}^{\text{N}}$ combination subsets of a fixed size N. Then, we conquer the combinatorial problem by assigning each combination a learnable score that is jointly optimized with its associate weights. We prove that the introduced scoring mechanism can well model the relative importance between combination subsets. And by gradually removing low-scored subsets, N:M fine-grained sparsity can be efficiently optimized during the normal training phase. Comprehensive experiments demonstrate that our learning best combination (LBC) performs consistently better than off-the-shelf N:M sparsity methods across various networks. Our code is released at \url{this https URL}.

60.The Open Kidney Ultrasound Data Set ⬇️

Ultrasound use is because of its low cost, non-ionizing, and non-invasive characteristics, and has established itself as a cornerstone radiological examination. Research on ultrasound applications has also expanded, especially with image analysis with machine learning. However, ultrasound data are frequently restricted to closed data sets, with only a few openly available. Despite being a frequently examined organ, the kidney lacks a publicly available ultrasonography data set. The proposed Open Kidney Ultrasound Data Set is the first publicly available set of kidney B-mode ultrasound data that includes annotations for multi-class semantic segmentation. It is based on data retrospectively collected in a 5-year period from over 500 patients with a mean age of 53.2 +/- 14.7 years, body mass index of 27.0 +/- 5.4 kg/m2, and most common primary diseases being diabetes mellitus, IgA nephropathy, and hypertension. There are labels for the view and fine-grained manual annotations from two expert sonographers. Notably, this data includes native and transplanted kidneys. Initial benchmarking measurements are performed, demonstrating a state-of-the-art algorithm achieving a Dice Sorenson Coefficient of 0.74 for the kidney capsule. This data set is a high-quality data set, including two sets of expert annotations, with a larger breadth of images than previously available. In increasing access to kidney ultrasound data, future researchers may be able to create novel image analysis techniques for tissue characterization, disease detection, and prognostication.

61.The Kidneys Are Not All Normal: Investigating the Speckle Distributions of Transplanted Kidneys ⬇️

Modelling ultrasound speckle has generated considerable interest for its ability to characterize tissue properties. As speckle is dependent on the underlying tissue architecture, modelling it may aid in tasks like segmentation or disease detection. However, for the transplanted kidney where ultrasound is commonly used to investigate dysfunction, it is currently unknown which statistical distribution best characterises such speckle. This is especially true for the regions of the transplanted kidney: the cortex, the medulla and the central echogenic complex. Furthermore, it is unclear how these distributions vary by patient variables such as age, sex, body mass index, primary disease, or donor type. These traits may influence speckle modelling given their influence on kidney anatomy. We are the first to investigate these two aims. N=821 kidney transplant recipient B-mode images were automatically segmented into the cortex, medulla, and central echogenic complex using a neural network. Seven distinct probability distributions were fitted to each region. The Rayleigh and Nakagami distributions had model parameters that differed significantly between the three regions (p <= 0.05). While both had excellent goodness of fit, the Nakagami had higher Kullbeck-Leibler divergence. Recipient age correlated weakly with scale in the cortex (Omega: rho = 0.11, p = 0.004), while body mass index correlated weakly with shape in the medulla (m: rho = 0.08, p = 0.04). Neither sex, primary disease, nor donor type demonstrated any correlation. We propose the Nakagami distribution be used to characterize transplanted kidneys regionally independent of disease etiology and most patient characteristics based on our findings.

62.ULTRA: Uncertainty-aware Label Distribution Learning for Breast Tumor Cellularity Assessment ⬇️

Neoadjuvant therapy (NAT) for breast cancer is a common treatment option in clinical practice. Tumor cellularity (TC), which represents the percentage of invasive tumors in the tumor bed, has been widely used to quantify the response of breast cancer to NAT. Therefore, automatic TC estimation is significant in clinical practice. However, existing state-of-the-art methods usually take it as a TC score regression problem, which ignores the ambiguity of TC labels caused by subjective assessment or multiple raters. In this paper, to efficiently leverage the label ambiguities, we proposed an Uncertainty-aware Label disTRibution leArning (ULTRA) framework for automatic TC estimation. The proposed ULTRA first converted the single-value TC labels to discrete label distributions, which effectively models the ambiguity among all possible TC labels. Furthermore, the network learned TC label distributions by minimizing the Kullback-Leibler (KL) divergence between the predicted and ground-truth TC label distributions, which better supervised the model to leverage the ambiguity of TC labels. Moreover, the ULTRA mimicked the multi-rater fusion process in clinical practice with a multi-branch feature fusion module to further explore the uncertainties of TC labels. We evaluated the ULTRA on the public BreastPathQ dataset. The experimental results demonstrate that the ULTRA outperformed the regression-based methods for a large margin and achieved state-of-the-art results. The code will be available from this https URL

63.CorticalFlow$^{++}$: Boosting Cortical Surface Reconstruction Accuracy, Regularity, and Interoperability ⬇️

The problem of Cortical Surface Reconstruction from magnetic resonance imaging has been traditionally addressed using lengthy pipelines of image processing techniques like FreeSurfer, CAT, or CIVET. These frameworks require very long runtimes deemed unfeasible for real-time applications and unpractical for large-scale studies. Recently, supervised deep learning approaches have been introduced to speed up this task cutting down the reconstruction time from hours to seconds. Using the state-of-the-art CorticalFlow model as a blueprint, this paper proposes three modifications to improve its accuracy and interoperability with existing surface analysis tools, while not sacrificing its fast inference time and low GPU memory consumption. First, we employ a more accurate ODE solver to reduce the diffeomorphic mapping approximation error. Second, we devise a routine to produce smoother template meshes avoiding mesh artifacts caused by sharp edges in CorticalFlow's convex-hull based template. Last, we recast pial surface prediction as the deformation of the predicted white surface leading to a one-to-one mapping between white and pial surface vertices. This mapping is essential to many existing surface analysis tools for cortical morphometry. We name the resulting method CorticalFlow$^{++}$. Using large-scale datasets, we demonstrate the proposed changes provide more geometric accuracy and surface regularity while keeping the reconstruction time and GPU memory requirements almost unchanged.

64.Physics Informed Neural Fields for Smoke Reconstruction with Sparse Data ⬇️

High-fidelity reconstruction of fluids from sparse multiview RGB videos remains a formidable challenge due to the complexity of the underlying physics as well as complex occlusion and lighting in captures. Existing solutions either assume knowledge of obstacles and lighting, or only focus on simple fluid scenes without obstacles or complex lighting, and thus are unsuitable for real-world scenes with unknown lighting or arbitrary obstacles. We present the first method to reconstruct dynamic fluid by leveraging the governing physics (ie, Navier -Stokes equations) in an end-to-end optimization from sparse videos without taking lighting conditions, geometry information, or boundary conditions as input. We provide a continuous spatio-temporal scene representation using neural networks as the ansatz of density and velocity solution functions for fluids as well as the radiance field for static objects. With a hybrid architecture that separates static and dynamic contents, fluid interactions with static obstacles are reconstructed for the first time without additional geometry input or human labeling. By augmenting time-varying neural radiance fields with physics-informed deep learning, our method benefits from the supervision of images and physical priors. To achieve robust optimization from sparse views, we introduced a layer-by-layer growing strategy to progressively increase the network capacity. Using progressively growing models with a new regularization term, we manage to disentangle density-color ambiguity in radiance fields without overfitting. A pretrained density-to-velocity fluid model is leveraged in addition as the data prior to avoid suboptimal velocity which underestimates vorticity but trivially fulfills physical equations. Our method exhibits high-quality results with relaxed constraints and strong flexibility on a representative set of synthetic and real flow captures.

65.Med-DANet: Dynamic Architecture Network for Efficient Medical Volumetric Segmentation ⬇️

For 3D medical image (e.g. CT and MRI) segmentation, the difficulty of segmenting each slice in a clinical case varies greatly. Previous research on volumetric medical image segmentation in a slice-by-slice manner conventionally use the identical 2D deep neural network to segment all the slices of the same case, ignoring the data heterogeneity among image slices. In this paper, we focus on multi-modal 3D MRI brain tumor segmentation and propose a dynamic architecture network named Med-DANet based on adaptive model selection to achieve effective accuracy and efficiency trade-off. For each slice of the input 3D MRI volume, our proposed method learns a slice-specific decision by the Decision Network to dynamically select a suitable model from the predefined Model Bank for the subsequent 2D segmentation task. Extensive experimental results on both BraTS 2019 and 2020 datasets show that our proposed method achieves comparable or better results than previous state-of-the-art methods for 3D MRI brain tumor segmentation with much less model complexity. Compared with the state-of-the-art 3D method TransBTS, the proposed framework improves the model efficiency by up to 3.5x without sacrificing the accuracy. Our code will be publicly available soon.

66.Safe Output Feedback Motion Planning from Images via Learned Perception Modules and Contraction Theory ⬇️

We present a motion planning algorithm for a class of uncertain control-affine nonlinear systems which guarantees runtime safety and goal reachability when using high-dimensional sensor measurements (e.g., RGB-D images) and a learned perception module in the feedback control loop. First, given a dataset of states and observations, we train a perception system that seeks to invert a subset of the state from an observation, and estimate an upper bound on the perception error which is valid with high probability in a trusted domain near the data. Next, we use contraction theory to design a stabilizing state feedback controller and a convergent dynamic state observer which uses the learned perception system to update its state estimate. We derive a bound on the trajectory tracking error when this controller is subjected to errors in the dynamics and incorrect state estimates. Finally, we integrate this bound into a sampling-based motion planner, guiding it to return trajectories that can be safely tracked at runtime using sensor data. We demonstrate our approach in simulation on a 4D car, a 6D planar quadrotor, and a 17D manipulation task with RGB(-D) sensor measurements, demonstrating that our method safely and reliably steers the system to the goal, while baselines that fail to consider the trusted domain or state estimation errors can be unsafe.

67.Pixel-by-pixel Mean Opinion Score (pMOS) for No-Reference Image Quality Assessment ⬇️

Deep-learning based techniques have contributed to the remarkable progress in the field of automatic image quality assessment (IQA). Existing IQA methods are designed to measure the quality of an image in terms of Mean Opinion Score (MOS) at the image-level (i.e. the whole image) or at the patch-level (dividing the image into multiple units and measuring quality of each patch). Some applications may require assessing the quality at the pixel-level (i.e. MOS value for each pixel), however, this is not possible in case of existing techniques as the spatial information is lost owing to their network structures. This paper proposes an IQA algorithm that can measure the MOS at the pixel-level, in addition to the image-level MOS. The proposed algorithm consists of three core parts, namely: i) Local IQA; ii) Region of Interest (ROI) prediction; iii) High-level feature embedding. The Local IQA part outputs the MOS at the pixel-level, or pixel-by-pixel MOS - we term it 'pMOS'. The ROI prediction part outputs weights that characterize the relative importance of region when calculating the image-level IQA. The high-level feature embedding part extracts high-level image features which are then embedded into the Local IQA part. In other words, the proposed algorithm yields three outputs: the pMOS which represents MOS for each pixel, the weights from the ROI indicating the relative importance of region, and finally the image-level MOS that is obtained by the weighted sum of pMOS and ROI values. The image-level MOS thus obtained by utilizing pMOS and ROI weights shows superior performance compared to the existing popular IQA techniques. In addition, visualization results indicate that predicted pMOS and ROI outputs are reasonably aligned with the general principles of the human visual system (HVS).

68.LST: Ladder Side-Tuning for Parameter and Memory Efficient Transfer Learning ⬇️

Fine-tuning large pre-trained models on downstream tasks has been adopted in a variety of domains recently. However, it is costly to update the entire parameter set of large pre-trained models. Although recently proposed parameter-efficient transfer learning (PETL) techniques allow updating a small subset of parameters (e.g. only using 2% of parameters) inside a pre-trained backbone network for a new task, they only reduce the training memory requirement by up to 30%. This is because the gradient computation for the trainable parameters still requires backpropagation through the large pre-trained backbone model. To address this, we propose Ladder Side-Tuning (LST), a new PETL technique that reduces training memory requirements by more substantial amounts. Unlike existing parameter-efficient methods that insert additional parameters inside backbone networks, we train a ladder side network, a small and separate network that takes intermediate activations as input via shortcut connections (ladders) from backbone networks and makes predictions. LST has significantly lower memory requirements than previous methods, because it does not require backpropagation through the backbone network, but instead only through the side network and ladder connections. We evaluate our method with various models (T5, CLIP-T5) on both NLP (GLUE) and vision-language (VQA, GQA, NLVR2, MSCOCO) tasks. LST saves 69% of the memory costs to fine-tune the whole network, while other methods only save 26% of that in similar parameter usages (hence, 2.7x more memory savings). Moreover, LST achieves higher accuracy than Adapter and LoRA in a low-memory regime. To further show the advantage of this better memory efficiency, we also apply LST to larger T5 models (T5-large, T5-3B), attaining better GLUE performance than full fine-tuning and other PETL methods. The exact same trend also holds in our experiments on VL tasks.

69.BEHAVIOR in Habitat 2.0: Simulator-Independent Logical Task Description for Benchmarking Embodied AI Agents ⬇️

Robots excel in performing repetitive and precision-sensitive tasks in controlled environments such as warehouses and factories, but have not been yet extended to embodied AI agents providing assistance in household tasks. Inspired by the catalyzing effect that benchmarks have played in the AI fields such as computer vision and natural language processing, the community is looking for new benchmarks for embodied AI. Prior work in embodied AI benchmark defines tasks using a different formalism, often specific to one environment, simulator or domain, making it hard to develop general and comparable solutions. In this work, we bring a subset of BEHAVIOR activities into Habitat 2.0 to benefit from its fast simulation speed, as a first step towards demonstrating the ease of adapting activities defined in the logic space into different simulators.

70.Assessing Privacy Leakage in Synthetic 3-D PET Imaging using Transversal GAN ⬇️

Training computer-vision related algorithms on medical images for disease diagnosis or image segmentation is difficult in large part due to privacy concerns. For this reason, generative image models are highly sought after to facilitate data sharing. However, 3-D generative models are understudied, and investigation of their privacy leakage is needed. We introduce our 3-D generative model, Transversal GAN (TrGAN), using head & neck PET images which are conditioned on tumour masks as a case study. We define quantitative measures of image fidelity, utility and privacy for our model. These metrics are evaluated in the course of training to identify ideal fidelity, utility and privacy trade-offs and establish the relationships between these parameters. We show that the discriminator of the TrGAN is vulnerable to attack, and that an attacker can identify which samples were used in training with almost perfect accuracy (AUC = 0.99). We also show that an attacker with access to only the generator cannot reliably classify whether a sample had been used for training (AUC = 0.51). This suggests that TrGAN generators, but not discriminators, may be used for sharing synthetic 3-D PET data with minimal privacy risk while maintaining good utility and fidelity.

71.Fitting Segmentation Networks on Varying Image Resolutions using Splatting ⬇️

Data used in image segmentation are not always defined on the same grid. This is particularly true for medical images, where the resolution, field-of-view and orientation can differ across channels and subjects. Images and labels are therefore commonly resampled onto the same grid, as a pre-processing step. However, the resampling operation introduces partial volume effects and blurring, thereby changing the effective resolution and reducing the contrast between structures. In this paper we propose a splat layer, which automatically handles resolution mismatches in the input data. This layer pushes each image onto a mean space where the forward pass is performed. As the splat operator is the adjoint to the resampling operator, the mean-space prediction can be pulled back to the native label space, where the loss function is computed. Thus, the need for explicit resolution adjustment using interpolation is removed. We show on two publicly available datasets, with simulated and real multi-modal magnetic resonance images, that this model improves segmentation results compared to resampling as a pre-processing step.