Computer Vision and Pattern Recognition (cs.CV)

Abstract: We aim to provide a general framework of for computational photography that recovers the real scene from imperfect images, via the Deep Nonparametric Convexified Filtering (DNCF). It is consists of a nonparametric deep network to resemble the physical equations behind the image formation, such as denoising, super-resolution, inpainting, and flash. DNCF has no parameterization dependent on training data, therefore has a strong generalization and robustness to adversarial image manipulation. During inference, we also encourage the network parameters to be nonnegative and create a bi-convex function on the input and parameters, and this adapts to second-order optimization algorithms with insufficient running time, having 10X acceleration over Deep Image Prior. With these tools, we empirically verify its capability to defend image classification deep networks against adversary attack algorithms in real-time.

Abstract: Audio-Visual Segmentation (AVS) aims to precisely outline audible objects in a visual scene at the pixel level. Existing AVS methods require fine-grained annotations of audio-mask pairs in supervised learning fashion. This limits their scalability since it is time consuming and tedious to acquire such cross-modality pixel level labels. To overcome this obstacle, in this work we introduce unsupervised audio-visual segmentation with no need for task-specific data annotations and model training. For tackling this newly proposed problem, we formulate a novel Cross-Modality Semantic Filtering (CMSF) approach to accurately associate the underlying audio-mask pairs by leveraging the off-the-shelf multi-modal foundation models (e.g., detection [1], open-world segmentation [2] and multi-modal alignment [3]). Guiding the proposal generation by either audio or visual cues, we design two training-free variants: AT-GDINO-SAM and OWOD-BIND. Extensive experiments on the AVS-Bench dataset show that our unsupervised approach can perform well in comparison to prior art supervised counterparts across complex scenarios with multiple auditory objects. Particularly, in situations where existing supervised AVS methods struggle with overlapping foreground objects, our models still excel in accurately segmenting overlapped auditory objects. Our code will be publicly released.

Abstract: High-resolution multi-modality information acquired by vision-based tactile sensors can support more dexterous manipulations for robot fingers. Optical flow is low-level information directly obtained by vision-based tactile sensors, which can be transformed into other modalities like force, geometry and depth. Current vision-tactile sensors employ optical flow methods from OpenCV to estimate the deformation of markers in gels. However, these methods need to be more precise for accurately measuring the displacement of markers during large elastic deformation of the gel, as this can significantly impact the accuracy of downstream tasks. This study proposes a self-supervised optical flow method based on deep learning to achieve high accuracy in displacement measurement for vision-based tactile sensors. The proposed method employs a coarse-to-fine strategy to handle large deformations by constructing a multi-scale feature pyramid from the input image. To better deal with the elastic deformation caused by the gel, the Helmholtz velocity decomposition constraint combined with the elastic deformation constraint are adopted to address the distortion rate and area change rate, respectively. A local flow fusion module is designed to smooth the optical flow, taking into account the prior knowledge of the blurred effect of gel deformation. We trained the proposed self-supervised network using an open-source dataset and compared it with traditional and deep learning-based optical flow methods. The results show that the proposed method achieved the highest displacement measurement accuracy, thereby demonstrating its potential for enabling more precise measurement of downstream tasks using vision-based tactile sensors.

Abstract: Autonomous driving systems require real-time environmental perception to ensure user safety and experience. Streaming perception is a task of reporting the current state of the world, which is used to evaluate the delay and accuracy of autonomous driving systems. In real-world applications, factors such as hardware limitations and high temperatures inevitably cause delays in autonomous driving systems, resulting in the offset between the model output and the world state. In order to solve this problem, this paper propose the Multi- Timestep Detector (MTD), an end-to-end detector which uses dynamic routing for multi-branch future prediction, giving model the ability to resist delay fluctuations. A Delay Analysis Module (DAM) is proposed to optimize the existing delay sensing method, continuously monitoring the model inference stack and calculating the delay trend. Moreover, a novel Timestep Branch Module (TBM) is constructed, which includes static flow and adaptive flow to adaptively predict specific timesteps according to the delay trend. The proposed method has been evaluated on the Argoverse-HD dataset, and the experimental results show that it has achieved state-of-the-art performance across various delay settings.

Abstract: Human affect recognition has been a significant topic in psychophysics and computer vision. However, the currently published datasets have many limitations. For example, most datasets contain frames that contain only information about facial expressions. Due to the limitations of previous datasets, it is very hard to either understand the mechanisms for affect recognition of humans or generalize well on common cases for computer vision models trained on those datasets. In this work, we introduce a brand new large dataset, the Video-based Emotion and Affect Tracking in Context Dataset (VEATIC), that can conquer the limitations of the previous datasets. VEATIC has 124 video clips from Hollywood movies, documentaries, and home videos with continuous valence and arousal ratings of each frame via real-time annotation. Along with the dataset, we propose a new computer vision task to infer the affect of the selected character via both context and character information in each video frame. Additionally, we propose a simple model to benchmark this new computer vision task. We also compare the performance of the pretrained model using our dataset with other similar datasets. Experiments show the competing results of our pretrained model via VEATIC, indicating the generalizability of VEATIC. Our dataset is available at https://veatic.github.io.

Abstract: In the domain of traffic safety and road maintenance, precise detection of road damage is crucial for ensuring safe driving and prolonging road durability. However, current methods often fall short due to limited data. Prior attempts have used Generative Adversarial Networks to generate damage with diverse shapes and manually integrate it into appropriate positions. However, the problem has not been well explored and is faced with two challenges. First, they only enrich the location and shape of damage while neglect the diversity of severity levels, and the realism still needs further improvement. Second, they require a significant amount of manual effort. To address these challenges, we propose an innovative approach. In addition to using GAN to generate damage with various shapes, we further employ texture synthesis techniques to extract road textures. These two elements are then mixed with different weights, allowing us to control the severity of the synthesized damage, which are then embedded back into the original images via Poisson blending. Our method ensures both richness of damage severity and a better alignment with the background. To save labor costs, we leverage structural similarity for automated sample selection during embedding. Each augmented data of an original image contains versions with varying severity levels. We implement a straightforward screening strategy to mitigate distribution drift. Experiments are conducted on a public road damage dataset. The proposed method not only eliminates the need for manual labor but also achieves remarkable enhancements, improving the mAP by 4.1% and the F1-score by 4.5%.

Abstract: Traffic signs are important facilities to ensure traffic safety and smooth flow, but may be damaged due to many reasons, which poses a great safety hazard. Therefore, it is important to study a method to detect damaged traffic signs. Existing object detection techniques for damaged traffic signs are still absent. Since damaged traffic signs are closer in appearance to normal ones, it is difficult to capture the detailed local damage features of damaged traffic signs using traditional object detection methods. In this paper, we propose an improved object detection method based on YOLOv5s, namely MFL-YOLO (Mutual Feature Levels Loss enhanced YOLO). We designed a simple cross-level loss function so that each level of the model has its own role, which is beneficial for the model to be able to learn more diverse features and improve the fine granularity. The method can be applied as a plug-and-play module and it does not increase the structural complexity or the computational complexity while improving the accuracy. We also replaced the traditional convolution and CSP with the GSConv and VoVGSCSP in the neck of YOLOv5s to reduce the scale and computational complexity. Compared with YOLOv5s, our MFL-YOLO improves 4.3 and 5.1 in F1 scores and mAP, while reducing the FLOPs by 8.9%. The Grad-CAM heat map visualization shows that our model can better focus on the local details of the damaged traffic signs. In addition, we also conducted experiments on CCTSDB2021 and TT100K to further validate the generalization of our model.

Abstract: Remote sensing object detection (RSOD), one of the most fundamental and challenging tasks in the remote sensing field, has received longstanding attention. In recent years, deep learning techniques have demonstrated robust feature representation capabilities and led to a big leap in the development of RSOD techniques. In this era of rapid technical evolution, this review aims to present a comprehensive review of the recent achievements in deep learning based RSOD methods. More than 300 papers are covered in this review. We identify five main challenges in RSOD, including multi-scale object detection, rotated object detection, weak object detection, tiny object detection, and object detection with limited supervision, and systematically review the corresponding methods developed in a hierarchical division manner. We also review the widely used benchmark datasets and evaluation metrics within the field of RSOD, as well as the application scenarios for RSOD. Future research directions are provided for further promoting the research in RSOD.

Abstract: For SLAM to be safely deployed in unstructured real world environments, it must possess several key properties that are not encompassed by conventional benchmarks. In this paper we show that SLAM commutativity, that is, consistency in trajectory estimates on forward and reverse traverses of the same route, is a significant issue for the state of the art. Current pipelines show a significant bias between forward and reverse directions of travel, that is in addition inconsistent regarding which direction of travel exhibits better performance. In this paper we propose several contributions to feature-based SLAM pipelines that remedies the motion bias problem. In a comprehensive evaluation across four datasets, we show that our contributions implemented in ORB-SLAM2 substantially reduce the bias between forward and backward motion and additionally improve the aggregated trajectory error. Removing the SLAM motion bias has significant relevance for the wide range of robotics and computer vision applications where performance consistency is important.

Abstract: The long-standing problem of novel view synthesis has many applications, notably in sports broadcasting. Photorealistic novel view synthesis of soccer actions, in particular, is of enormous interest to the broadcast industry. Yet only a few industrial solutions have been proposed, and even fewer that achieve near-broadcast quality of the synthetic replays. Except for their setup of multiple static cameras around the playfield, the best proprietary systems disclose close to no information about their inner workings. Leveraging multiple static cameras for such a task indeed presents a challenge rarely tackled in the literature, for a lack of public datasets: the reconstruction of a large-scale, mostly static environment, with small, fast-moving elements. Recently, the emergence of neural radiance fields has induced stunning progress in many novel view synthesis applications, leveraging deep learning principles to produce photorealistic results in the most challenging settings. In this work, we investigate the feasibility of basing a solution to the task on dynamic NeRFs, i.e., neural models purposed to reconstruct general dynamic content. We compose synthetic soccer environments and conduct multiple experiments using them, identifying key components that help reconstruct soccer scenes with dynamic NeRFs. We show that, although this approach cannot fully meet the quality requirements for the target application, it suggests promising avenues toward a cost-efficient, automatic solution. We also make our work dataset and code publicly available, with the goal to encourage further efforts from the research community on the task of novel view synthesis for dynamic soccer scenes. For code, data, and video results, please see https://soccernerfs.isach.be.

Abstract: While several previous studies have devised methods for segmentation of polyps, most of these methods are not rigorously assessed on multi-center datasets. Variability due to appearance of polyps from one center to another, difference in endoscopic instrument grades, and acquisition quality result in methods with good performance on in-distribution test data, and poor performance on out-of-distribution or underrepresented samples. Unfair models have serious implications and pose a critical challenge to clinical applications. We adapt an implicit bias mitigation method which leverages Bayesian epistemic uncertainties during training to encourage the model to focus on underrepresented sample regions. We demonstrate the potential of this approach to improve generalisability without sacrificing state-of-the-art performance on a challenging multi-center polyp segmentation dataset (PolypGen) with different centers and image modalities.

Abstract: Vision and Language Models (VLMs), such as CLIP, have enabled visual recognition of a potentially unlimited set of categories described by text prompts. However, for the best visual recognition performance, these models still require tuning to better fit the data distributions of the downstream tasks, in order to overcome the domain shift from the web-based pre-training data. Recently, it has been shown that it is possible to effectively tune VLMs without any paired data, and in particular to effectively improve VLMs visual recognition performance using text-only training data generated by Large Language Models (LLMs). In this paper, we dive deeper into this exciting text-only VLM training approach and explore ways it can be significantly further improved taking the specifics of the downstream task into account when sampling text data from LLMs. In particular, compared to the SOTA text-only VLM training approach, we demonstrate up to 8.4% performance improvement in (cross) domain-specific adaptation, up to 8.7% improvement in fine-grained recognition, and 3.1% overall average improvement in zero-shot classification compared to strong baselines.

Abstract: Shape assembly aims to reassemble parts (or fragments) into a complete object, which is a common task in our daily life. Different from the semantic part assembly (e.g., assembling a chair's semantic parts like legs into a whole chair), geometric part assembly (e.g., assembling bowl fragments into a complete bowl) is an emerging task in computer vision and robotics. Instead of semantic information, this task focuses on geometric information of parts. As the both geometric and pose space of fractured parts are exceptionally large, shape pose disentanglement of part representations is beneficial to geometric shape assembly. In our paper, we propose to leverage SE(3) equivariance for such shape pose disentanglement. Moreover, while previous works in vision and robotics only consider SE(3) equivariance for the representations of single objects, we move a step forward and propose leveraging SE(3) equivariance for representations considering multi-part correlations, which further boosts the performance of the multi-part assembly. Experiments demonstrate the significance of SE(3) equivariance and our proposed method for geometric shape assembly. Project page: https://crtie.github.io/SE-3-part-assembly/

Abstract: Early detection and tracking of ejecta in the vicinity of small solar system bodies is crucial to guarantee spacecraft safety and support scientific observation. During the visit of active asteroid Bennu, the OSIRIS-REx spacecraft relied on the analysis of images captured by onboard navigation cameras to detect particle ejection events, which ultimately became one of the mission's scientific highlights. To increase the scientific return of similar time-constrained missions, this work proposes an event-based solution that is dedicated to the detection and tracking of centimetre-sized particles. Unlike a standard frame-based camera, the pixels of an event-based camera independently trigger events indicating whether the scene brightness has increased or decreased at that time and location in the sensor plane. As a result of the sparse and asynchronous spatiotemporal output, event cameras combine very high dynamic range and temporal resolution with low-power consumption, which could complement existing onboard imaging techniques. This paper motivates the use of a scientific event camera by reconstructing the particle ejection episodes reported by the OSIRIS-REx mission in a photorealistic scene generator and in turn, simulating event-based observations. The resulting streams of spatiotemporal data support future work on event-based multi-object tracking.

Abstract: Segment anything model (SAM), an eminent universal image segmentation model, has recently gathered considerable attention within the domain of medical image segmentation. Despite the remarkable performance of SAM on natural images, it grapples with significant performance degradation and limited generalization when confronted with medical images, particularly with those involving objects of low contrast, faint boundaries, intricate shapes, and diminutive sizes. In this paper, we propose SAMUS, a universal model tailored for ultrasound image segmentation. In contrast to previous SAM-based universal models, SAMUS pursues not only better generalization but also lower deployment cost, rendering it more suitable for clinical applications. Specifically, based on SAM, a parallel CNN branch is introduced to inject local features into the ViT encoder through cross-branch attention for better medical image segmentation. Then, a position adapter and a feature adapter are developed to adapt SAM from natural to medical domains and from requiring large-size inputs (1024x1024) to small-size inputs (256x256) for more clinical-friendly deployment. A comprehensive ultrasound dataset, comprising about 30k images and 69k masks and covering six object categories, is collected for verification. Extensive comparison experiments demonstrate SAMUS's superiority against the state-of-the-art task-specific models and universal foundation models under both task-specific evaluation and generalization evaluation. Moreover, SAMUS is deployable on entry-level GPUs, as it has been liberated from the constraints of long sequence encoding. The code, data, and models will be released at https://github.com/xianlin7/SAMUS.

Abstract: Magnetic resonance imaging~(MRI) have played a crucial role in brain disease diagnosis, with which a range of computer-aided artificial intelligence methods have been proposed. However, the early explorations usually focus on the limited types of brain diseases in one study and train the model on the data in a small scale, yielding the bottleneck of generalization. Towards a more effective and scalable paradigm, we propose a hierarchical knowledge-enhanced pre-training framework for the universal brain MRI diagnosis, termed as UniBrain. Specifically, UniBrain leverages a large-scale dataset of 24,770 imaging-report pairs from routine diagnostics. Different from previous pre-training techniques for the unitary vision or textual feature, or with the brute-force alignment between vision and language information, we leverage the unique characteristic of report information in different granularity to build a hierarchical alignment mechanism, which strengthens the efficiency in feature learning. Our UniBrain is validated on three real world datasets with severe class imbalance and the public BraTS2019 dataset. It not only consistently outperforms all state-of-the-art diagnostic methods by a large margin and provides a superior grounding performance but also shows comparable performance compared to expert radiologists on certain disease types.

Abstract: The self-media era provides us tremendous high quality videos. Unfortunately, frequent video copyright infringements are now seriously damaging the interests and enthusiasm of video creators. Identifying infringing videos is therefore a compelling task. Current state-of-the-art methods tend to simply feed high-dimensional mixed video features into deep neural networks and count on the networks to extract useful representations. Despite its simplicity, this paradigm heavily relies on the original entangled features and lacks constraints guaranteeing that useful task-relevant semantics are extracted from the features. In this paper, we seek to tackle the above challenges from two aspects: (1) We propose to disentangle an original high-dimensional feature into multiple sub-features, explicitly disentangling the feature into exclusive lower-dimensional components. We expect the sub-features to encode non-overlapping semantics of the original feature and remove redundant information. (2) On top of the disentangled sub-features, we further learn an auxiliary feature to enhance the sub-features. We theoretically analyzed the mutual information between the label and the disentangled features, arriving at a loss that maximizes the extraction of task-relevant information from the original feature. Extensive experiments on two large-scale benchmark datasets (i.e., SVD and VCSL) demonstrate that our method achieves 90.1% TOP-100 mAP on the large-scale SVD dataset and also sets the new state-of-the-art on the VCSL benchmark dataset. Our code and model have been released at https://github.com/yyyooooo/DMI/, hoping to contribute to the community.

Abstract: Manufacturing industries require efficient and voluminous production of high-quality finished goods. In the context of Industry 4.0, visual anomaly detection poses an optimistic solution for automatically controlling product quality with high precision. Automation based on computer vision poses a promising solution to prevent bottlenecks at the product quality checkpoint. We considered recent advancements in machine learning to improve visual defect localization, but challenges persist in obtaining a balanced feature set and database of the wide variety of defects occurring in the production line. This paper proposes a defect localizing autoencoder with unsupervised class selection by clustering with k-means the features extracted from a pre-trained VGG-16 network. The selected classes of defects are augmented with natural wild textures to simulate artificial defects. The study demonstrates the effectiveness of the defect localizing autoencoder with unsupervised class selection for improving defect detection in manufacturing industries. The proposed methodology shows promising results with precise and accurate localization of quality defects on melamine-faced boards for the furniture industry. Incorporating artificial defects into the training data shows significant potential for practical implementation in real-world quality control scenarios.

Abstract: Convolutional networks and vision transformers have different forms of pairwise interactions, pooling across layers and pooling at the end of the network. Does the latter really need to be different? As a by-product of pooling, vision transformers provide spatial attention for free, but this is most often of low quality unless self-supervised, which is not well studied. Is supervision really the problem? In this work, we develop a generic pooling framework and then we formulate a number of existing methods as instantiations. By discussing the properties of each group of methods, we derive SimPool, a simple attention-based pooling mechanism as a replacement of the default one for both convolutional and transformer encoders. We find that, whether supervised or self-supervised, this improves performance on pre-training and downstream tasks and provides attention maps delineating object boundaries in all cases. One could thus call SimPool universal. To our knowledge, we are the first to obtain attention maps in supervised transformers of at least as good quality as self-supervised, without explicit losses or modifying the architecture. Code at: https://github.com/billpsomas/simpool.

Abstract: Large-scale text-to-image models including Stable Diffusion are capable of generating high-fidelity photorealistic portrait images. There is an active research area dedicated to personalizing these models, aiming to synthesize specific subjects or styles using provided sets of reference images. However, despite the plausible results from these personalization methods, they tend to produce images that often fall short of realism and are not yet on a commercially viable level. This is particularly noticeable in portrait image generation, where any unnatural artifact in human faces is easily discernible due to our inherent human bias. To address this, we introduce MagiCapture, a personalization method for integrating subject and style concepts to generate high-resolution portrait images using just a few subject and style references. For instance, given a handful of random selfies, our fine-tuned model can generate high-quality portrait images in specific styles, such as passport or profile photos. The main challenge with this task is the absence of ground truth for the composed concepts, leading to a reduction in the quality of the final output and an identity shift of the source subject. To address these issues, we present a novel Attention Refocusing loss coupled with auxiliary priors, both of which facilitate robust learning within this weakly supervised learning setting. Our pipeline also includes additional post-processing steps to ensure the creation of highly realistic outputs. MagiCapture outperforms other baselines in both quantitative and qualitative evaluations and can also be generalized to other non-human objects.

Abstract: Traffic sign detection is an important research direction in intelligent driving. Unfortunately, existing methods often overlook extreme conditions such as fog, rain, and motion blur. Moreover, the end-to-end training strategy for image denoising and object detection models fails to utilize inter-model information effectively. To address these issues, we propose CCSPNet, an efficient feature extraction module based on Transformers and CNNs, which effectively leverages contextual information, achieves faster inference speed and provides stronger feature enhancement capabilities. Furthermore, we establish the correlation between object detection and image denoising tasks and propose a joint training model, CCSPNet-Joint, to improve data efficiency and generalization. Finally, to validate our approach, we create the CCTSDB-AUG dataset for traffic sign detection in extreme scenarios. Extensive experiments have shown that CCSPNet achieves state-of-the-art performance in traffic sign detection under extreme conditions. Compared to end-to-end methods, CCSPNet-Joint achieves a 5.32% improvement in precision and an 18.09% improvement in [email protected].

Abstract: The recent surge in large-scale foundation models has spurred the development of efficient methods for adapting these models to various downstream tasks. Low-rank adaptation methods, such as LoRA, have gained significant attention due to their outstanding parameter efficiency and no additional inference latency. This paper investigates a more general form of adapter module based on the analysis that parallel and sequential adaptation branches learn novel and general features during fine-tuning, respectively. The proposed method, named Hydra, due to its multi-head computational branches, combines parallel and sequential branch to integrate capabilities, which is more expressive than existing single branch methods and enables the exploration of a broader range of optimal points in the fine-tuning process. In addition, the proposed adaptation method explicitly leverages the pre-trained weights by performing a linear combination of the pre-trained features. It allows the learned features to have better generalization performance across diverse downstream tasks. Furthermore, we perform a comprehensive analysis of the characteristics of each adaptation branch with empirical evidence. Through an extensive range of experiments, encompassing comparisons and ablation studies, we substantiate the efficiency and demonstrate the superior performance of Hydra. This comprehensive evaluation underscores the potential impact and effectiveness of Hydra in a variety of applications. Our code is available on \url{https://github.com/extremebird/Hydra}

Abstract: Video-based remote physiological measurement utilizes facial videos to measure the blood volume change signal, which is also called remote photoplethysmography (rPPG). Supervised methods for rPPG measurements have been shown to achieve good performance. However, the drawback of these methods is that they require facial videos with ground truth (GT) physiological signals, which are often costly and difficult to obtain. In this paper, we propose Contrast-Phys+, a method that can be trained in both unsupervised and weakly-supervised settings. We employ a 3DCNN model to generate multiple spatiotemporal rPPG signals and incorporate prior knowledge of rPPG into a contrastive loss function. We further incorporate the GT signals into contrastive learning to adapt to partial or misaligned labels. The contrastive loss encourages rPPG/GT signals from the same video to be grouped together, while pushing those from different videos apart. We evaluate our methods on five publicly available datasets that include both RGB and Near-infrared videos. Contrast-Phys+ outperforms the state-of-the-art supervised methods, even when using partially available or misaligned GT signals, or no labels at all. Additionally, we highlight the advantages of our methods in terms of computational efficiency, noise robustness, and generalization.

Abstract: Recent progresses in large-scale text-to-image models have yielded remarkable accomplishments, finding various applications in art domain. However, expressing unique characteristics of an artwork (e.g. brushwork, colortone, or composition) with text prompts alone may encounter limitations due to the inherent constraints of verbal description. To this end, we introduce DreamStyler, a novel framework designed for artistic image synthesis, proficient in both text-to-image synthesis and style transfer. DreamStyler optimizes a multi-stage textual embedding with a context-aware text prompt, resulting in prominent image quality. In addition, with content and style guidance, DreamStyler exhibits flexibility to accommodate a range of style references. Experimental results demonstrate its superior performance across multiple scenarios, suggesting its promising potential in artistic product creation.

Abstract: The goal of low-light image enhancement is to restore the color and details of the image and is of great significance for high-level visual tasks in autonomous driving. However, it is difficult to restore the lost details in the dark area by relying only on the RGB domain. In this paper we introduce frequency as a new clue into the network and propose a novel DCT-driven enhancement transformer (DEFormer). First, we propose a learnable frequency branch (LFB) for frequency enhancement contains DCT processing and curvature-based frequency enhancement (CFE). CFE calculates the curvature of each channel to represent the detail richness of different frequency bands, then we divides the frequency features, which focuses on frequency bands with richer textures. In addition, we propose a cross domain fusion (CDF) for reducing the differences between the RGB domain and the frequency domain. We also adopt DEFormer as a preprocessing in dark detection, DEFormer effectively improves the performance of the detector, bringing 2.1% and 3.4% improvement in ExDark and DARK FACE datasets on mAP respectively.

Abstract: Human action recognition (HAR) is a high-level and significant research area in computer vision due to its ubiquitous applications. The main limitations of the current HAR models are their complex structures and lengthy training time. In this paper, we propose a simple yet versatile and effective end-to-end deep learning architecture, coined as TransNet, for HAR. TransNet decomposes the complex 3D-CNNs into 2D- and 1D-CNNs, where the 2D- and 1D-CNN components extract spatial features and temporal patterns in videos, respectively. Benefiting from its concise architecture, TransNet is ideally compatible with any pretrained state-of-the-art 2D-CNN models in other fields, being transferred to serve the HAR task. In other words, it naturally leverages the power and success of transfer learning for HAR, bringing huge advantages in terms of efficiency and effectiveness. Extensive experimental results and the comparison with the state-of-the-art models demonstrate the superior performance of the proposed TransNet in HAR in terms of flexibility, model complexity, training speed and classification accuracy.

Abstract: Background. Cardiac dominance classification is essential for SYNTAX score estimation, which is a tool used to determine the complexity of coronary artery disease and guide patient selection toward optimal revascularization strategy. Objectives. Cardiac dominance classification algorithm based on the analysis of right coronary artery (RCA) angiograms using neural network Method. We employed convolutional neural network ConvNext and Swin transformer for 2D image (frames) classification, along with a majority vote for cardio angiographic view classification. An auxiliary network was also used to detect irrelevant images which were then excluded from the data set. Our data set consisted of 828 angiographic studies, 192 of them being patients with left dominance. Results. 5-fold cross validation gave the following dominance classification metrics (p=95%): macro recall=93.1%, accuracy=93.5%, macro F1=89.2%. The most common case in which the model regularly failed was RCA occlusion, as it requires utilization of LCA information. Another cause for false prediction is a small diameter combined with poor quality cardio angiographic view. In such cases, cardiac dominance classification can be complex and may require discussion among specialists to reach an accurate conclusion. Conclusion. The use of machine learning approaches to classify cardiac dominance based on RCA alone has been shown to be successful with satisfactory accuracy. However, for higher accuracy, it is necessary to utilize LCA information in the case of an occluded RCA and detect cases where there is high uncertainty.

Abstract: Benchmark datasets for digital dermatology unwittingly contain inaccuracies that reduce trust in model performance estimates. We propose a resource-efficient data cleaning protocol to identify issues that escaped previous curation. The protocol leverages an existing algorithmic cleaning strategy and is followed by a confirmation process terminated by an intuitive stopping criterion. Based on confirmation by multiple dermatologists, we remove irrelevant samples and near duplicates and estimate the percentage of label errors in six dermatology image datasets for model evaluation promoted by the International Skin Imaging Collaboration. Along with this paper, we publish revised file lists for each dataset which should be used for model evaluation. Our work paves the way for more trustworthy performance assessment in digital dermatology.

Abstract: JPEG remains one of the most widespread lossy image coding methods. However, the non-differentiable nature of JPEG restricts the application in deep learning pipelines. Several differentiable approximations of JPEG have recently been proposed to address this issue. This paper conducts a comprehensive review of existing diff. JPEG approaches and identifies critical details that have been missed by previous methods. To this end, we propose a novel diff. JPEG approach, overcoming previous limitations. Our approach is differentiable w.r.t. the input image, the JPEG quality, the quantization tables, and the color conversion parameters. We evaluate the forward and backward performance of our diff. JPEG approach against existing methods. Additionally, extensive ablations are performed to evaluate crucial design choices. Our proposed diff. JPEG resembles the (non-diff.) reference implementation best, significantly surpassing the recent-best diff. approach by $3.47$dB (PSNR) on average. For strong compression rates, we can even improve PSNR by $9.51$dB. Strong adversarial attack results are yielded by our diff. JPEG, demonstrating the effective gradient approximation. Our code is available at https://github.com/necla-ml/Diff-JPEG.

Abstract: Generalized zero-shot learning(GZSL) aims to classify samples from seen and unseen labels, assuming unseen labels are not accessible during training. Recent advancements in GZSL have been expedited by incorporating contrastive-learning-based (instance-based) embedding in generative networks and leveraging the semantic relationship between data points. However, existing embedding architectures suffer from two limitations: (1) limited discriminability of synthetic features' embedding without considering fine-grained cluster structures; (2) inflexible optimization due to restricted scaling mechanisms on existing contrastive embedding networks, leading to overlapped representations in the embedding space. To enhance the quality of representations in the embedding space, as mentioned in (1), we propose a margin-based prototypical contrastive learning embedding network that reaps the benefits of prototype-data (cluster quality enhancement) and implicit data-data (fine-grained representations) interaction while providing substantial cluster supervision to the embedding network and the generator. To tackle (2), we propose an instance adaptive contrastive loss that leads to generalized representations for unseen labels with increased inter-class margin. Through comprehensive experimental evaluation, we show that our method can outperform the current state-of-the-art on three benchmark datasets. Our approach also consistently achieves the best unseen performance in the GZSL setting.

Abstract: Neural implicit modeling permits to achieve impressive 3D reconstruction results on small objects, while it exhibits significant limitations in large indoor scenes. In this work, we propose a novel neural implicit modeling method that leverages multiple regularization strategies to achieve better reconstructions of large indoor environments, while relying only on images. A sparse but accurate depth prior is used to anchor the scene to the initial model. A dense but less accurate depth prior is also introduced, flexible enough to still let the model diverge from it to improve the estimated geometry. Then, a novel self-supervised strategy to regularize the estimated surface normals is presented. Finally, a learnable exposure compensation scheme permits to cope with challenging lighting conditions. Experimental results show that our approach produces state-of-the-art 3D reconstructions in challenging indoor scenarios.

Abstract: Video deblurring methods, aiming at recovering consecutive sharp frames from a given blurry video, usually assume that the input video suffers from consecutively blurry frames. However, in real-world blurry videos taken by modern imaging devices, sharp frames usually appear in the given video, thus making temporal long-term sharp features available for facilitating the restoration of a blurry frame. In this work, we propose a video deblurring method that leverages both neighboring frames and present sharp frames using hybrid Transformers for feature aggregation. Specifically, we first train a blur-aware detector to distinguish between sharp and blurry frames. Then, a window-based local Transformer is employed for exploiting features from neighboring frames, where cross attention is beneficial for aggregating features from neighboring frames without explicit spatial alignment. To aggregate long-term sharp features from detected sharp frames, we utilize a global Transformer with multi-scale matching capability. Moreover, our method can easily be extended to event-driven video deblurring by incorporating an event fusion module into the global Transformer. Extensive experiments on benchmark datasets demonstrate that our proposed method outperforms state-of-the-art video deblurring methods as well as event-driven video deblurring methods in terms of quantitative metrics and visual quality. The source code and trained models are available at https://github.com/shangwei5/STGTN.

Abstract: Image reconstruction-based anomaly detection models are widely explored in industrial visual inspection. However, existing models usually suffer from the trade-off between normal reconstruction fidelity and abnormal reconstruction distinguishability, which damages the performance. In this paper, we find that the above trade-off can be better mitigated by leveraging the distinct frequency biases between normal and abnormal reconstruction errors. To this end, we propose Frequency-aware Image Restoration (FAIR), a novel self-supervised image restoration task that restores images from their high-frequency components. It enables precise reconstruction of normal patterns while mitigating unfavorable generalization to anomalies. Using only a simple vanilla UNet, FAIR achieves state-of-the-art performance with higher efficiency on various defect detection datasets. Code: https://github.com/liutongkun/FAIR.

Abstract: In this paper, we propose a novel training strategy called SupFusion, which provides an auxiliary feature level supervision for effective LiDAR-Camera fusion and significantly boosts detection performance. Our strategy involves a data enhancement method named Polar Sampling, which densifies sparse objects and trains an assistant model to generate high-quality features as the supervision. These features are then used to train the LiDAR-Camera fusion model, where the fusion feature is optimized to simulate the generated high-quality features. Furthermore, we propose a simple yet effective deep fusion module, which contiguously gains superior performance compared with previous fusion methods with SupFusion strategy. In such a manner, our proposal shares the following advantages. Firstly, SupFusion introduces auxiliary feature-level supervision which could boost LiDAR-Camera detection performance without introducing extra inference costs. Secondly, the proposed deep fusion could continuously improve the detector's abilities. Our proposed SupFusion and deep fusion module is plug-and-play, we make extensive experiments to demonstrate its effectiveness. Specifically, we gain around 2% 3D mAP improvements on KITTI benchmark based on multiple LiDAR-Camera 3D detectors.

Abstract: Objective: Neoadjuvant chemotherapy (NACT) is one kind of treatment for advanced stage ovarian cancer patients. However, due to the nature of tumor heterogeneity, the patients' responses to NACT varies significantly among different subgroups. To address this clinical challenge, the purpose of this study is to develop a novel image marker to achieve high accuracy response prediction of the NACT at an early stage. Methods: For this purpose, we first computed a total of 1373 radiomics features to quantify the tumor characteristics, which can be grouped into three categories: geometric, intensity, and texture features. Second, all these features were optimized by principal component analysis algorithm to generate a compact and informative feature cluster. Using this cluster as the input, an SVM based classifier was developed and optimized to create a final marker, indicating the likelihood of the patient being responsive to the NACT treatment. To validate this scheme, a total of 42 ovarian cancer patients were retrospectively collected. A nested leave-one-out cross-validation was adopted for model performance assessment. Results: The results demonstrate that the new method yielded an AUC (area under the ROC [receiver characteristic operation] curve) of 0.745. Meanwhile, the model achieved overall accuracy of 76.2%, positive predictive value of 70%, and negative predictive value of 78.1%. Conclusion: This study provides meaningful information for the development of radiomics based image markers in NACT response prediction.

Abstract: Monocular 3D object detection is a challenging task because depth information is difficult to obtain from 2D images. A subset of viewpoint-agnostic monocular 3D detection methods also do not explicitly leverage scene homography or geometry during training, meaning that a model trained thusly can detect objects in images from arbitrary viewpoints. Such works predict the projections of the 3D bounding boxes on the image plane to estimate the location of the 3D boxes, but these projections are not rectangular so the calculation of IoU between these projected polygons is not straightforward. This work proposes an efficient, fully differentiable algorithm for the calculation of IoU between two convex polygons, which can be utilized to compute the IoU between two 3D bounding box footprints viewed from an arbitrary angle. We test the performance of the proposed polygon IoU loss (PIoU loss) on three state-of-the-art viewpoint-agnostic 3D detection models. Experiments demonstrate that the proposed PIoU loss converges faster than L1 loss and that in 3D detection models, a combination of PIoU loss and L1 loss gives better results than L1 loss alone (+1.64% AP70 for MonoCon on cars, +0.18% AP70 for RTM3D on cars, and +0.83%/+2.46% AP50/AP25 for MonoRCNN on cyclists).

Abstract: RGB-D object recognition systems improve their predictive performances by fusing color and depth information, outperforming neural network architectures that rely solely on colors. While RGB-D systems are expected to be more robust to adversarial examples than RGB-only systems, they have also been proven to be highly vulnerable. Their robustness is similar even when the adversarial examples are generated by altering only the original images' colors. Different works highlighted the vulnerability of RGB-D systems; however, there is a lacking of technical explanations for this weakness. Hence, in our work, we bridge this gap by investigating the learned deep representation of RGB-D systems, discovering that color features make the function learned by the network more complex and, thus, more sensitive to small perturbations. To mitigate this problem, we propose a defense based on a detection mechanism that makes RGB-D systems more robust against adversarial examples. We empirically show that this defense improves the performances of RGB-D systems against adversarial examples even when they are computed ad-hoc to circumvent this detection mechanism, and that is also more effective than adversarial training.

Abstract: Deep neural network models can learn clinically relevant features from millions of histopathology images. However generating high-quality annotations to train such models for each hospital, each cancer type, and each diagnostic task is prohibitively laborious. On the other hand, terabytes of training data -- while lacking reliable annotations -- are readily available in the public domain in some cases. In this work, we explore how these large datasets can be consciously utilized to pre-train deep networks to encode informative representations. We then fine-tune our pre-trained models on a fraction of annotated training data to perform specific downstream tasks. We show that our approach can reach the state-of-the-art (SOTA) for patch-level classification with only 1-10% randomly selected annotations compared to other SOTA approaches. Moreover, we propose an uncertainty-aware loss function, to quantify the model confidence during inference. Quantified uncertainty helps experts select the best instances to label for further training. Our uncertainty-aware labeling reaches the SOTA with significantly fewer annotations compared to random labeling. Last, we demonstrate how our pre-trained encoders can surpass current SOTA for whole-slide image classification with weak supervision. Our work lays the foundation for data and task-agnostic pre-trained deep networks with quantified uncertainty.

Abstract: We study inferring a tree-structured representation from a single image for object shading. Prior work typically uses the parametric or measured representation to model shading, which is neither interpretable nor easily editable. We propose using the shade tree representation, which combines basic shading nodes and compositing methods to factorize object surface shading. The shade tree representation enables novice users who are unfamiliar with the physical shading process to edit object shading in an efficient and intuitive manner. A main challenge in inferring the shade tree is that the inference problem involves both the discrete tree structure and the continuous parameters of the tree nodes. We propose a hybrid approach to address this issue. We introduce an auto-regressive inference model to generate a rough estimation of the tree structure and node parameters, and then we fine-tune the inferred shade tree through an optimization algorithm. We show experiments on synthetic images, captured reflectance, real images, and non-realistic vector drawings, allowing downstream applications such as material editing, vectorized shading, and relighting. Project website: https://chen-geng.com/inv-shade-trees

Abstract: Our goal is to create a realistic 3D facial avatar with hair and accessories using only a text description. While this challenge has attracted significant recent interest, existing methods either lack realism, produce unrealistic shapes, or do not support editing, such as modifications to the hairstyle. We argue that existing methods are limited because they employ a monolithic modeling approach, using a single representation for the head, face, hair, and accessories. Our observation is that the hair and face, for example, have very different structural qualities that benefit from different representations. Building on this insight, we generate avatars with a compositional model, in which the head, face, and upper body are represented with traditional 3D meshes, and the hair, clothing, and accessories with neural radiance fields (NeRF). The model-based mesh representation provides a strong geometric prior for the face region, improving realism while enabling editing of the person's appearance. By using NeRFs to represent the remaining components, our method is able to model and synthesize parts with complex geometry and appearance, such as curly hair and fluffy scarves. Our novel system synthesizes these high-quality compositional avatars from text descriptions. The experimental results demonstrate that our method, Text-guided generation and Editing of Compositional Avatars (TECA), produces avatars that are more realistic than those of recent methods while being editable because of their compositional nature. For example, our TECA enables the seamless transfer of compositional features like hairstyles, scarves, and other accessories between avatars. This capability supports applications such as virtual try-on.