WACV 2024 论文列表

Understanding of human visual perception has historically inspired the design of computer vision architectures. As an example, perception occurs at different scales both spatially and temporally, suggesting that the extraction of salient visual information may be made more effective by attending to specific features at varying scales. Visual changes in the body due to physiological processes also occur at varying scales and with modality-specific characteristic properties. Inspired by this, we present BigSmall, an efficient architecture for physiological and behavioral measurement. We present the first joint camera-based facial action, cardiac, and pulmonary measurement model. We propose a multi-branch network with wrapping temporal shift modules that yields both accuracy and efficiency gains. We observe that fusing low-level features leads to suboptimal performance, but that fusing high level features enables efficiency gains with negligible losses in accuracy. Experimental results demonstrate that BigSmall significantly reduces the computational costs. Furthermore, compared to existing task-specific models, BigSmall achieves comparable or better results on multiple physiological measurement tasks simultaneously with a unified model.

Synthetic data is emerging as a substitute for authentic data to solve ethical and legal challenges in handling authentic face data. The current models can create real-looking face images of people who do not exist. However, it is a known and sensitive problem that face recognition systems are susceptible to bias, i.e. performance differences between different demographic and non-demographics attributes, which can lead to unfair decisions. In this work, we investigate how the diversity of synthetic face recognition datasets compares to authentic datasets, and how the distribution of the training data of the generative models affects the distribution of the synthetic data. To do this, we looked at the distribution of gender, ethnicity, age, and head position. Furthermore, we investigated the concrete bias of three recent synthetic-based face recognition models on the studied attributes in comparison to a baseline model trained on authentic data. Our results show that the generator generate a similar distribution as the used training data in terms of the different attributes. With regard to bias, it can be seen that the synthetic-based models share a similar bias behavior with the authentic-based models. However, with the uncovered lower intra-identity attribute consistency seems to be beneficial in reducing bias.

Composed image retrieval searches for a target image based on a multi-modal user query comprised of a reference image and modification text describing the desired changes. Existing approaches to solving this challenging task learn a mapping from the (reference image, modification text)-pair to an image embedding that is then matched against a large image corpus. One area that has not yet been explored is the reverse direction, which asks the question, what reference image when modified as described by the text would produce the given target image? In this work we propose a bi-directional training scheme that leverages such reversed queries and can be applied to existing composed image retrieval architectures with minimum changes, which improves the performance of the model. To encode the bi-directional query we prepend a learnable token to the modification text that designates the direction of the query and then finetune the parameters of the text embedding module. We make no other changes to the network architecture. Experiments on two standard datasets show that our novel approach achieves improved performance over a baseline BLIP-based model that itself already achieves competitive performance. Our code is released at https://github.com/Cuberick-Orion/Bi-Blip4CIR.

Medical image segmentation has seen significant improvements with transformer models, which excel in grasping far-reaching contexts and global contextual information. However, the increasing computational demands of these models, proportional to the squared token count, limit their depth and resolution capabilities. Most current methods process D volumetric image data slice-by-slice (called pseudo 3D), missing crucial inter-slice information and thus reducing the model's overall performance. To address these challenges, we introduce the concept of Deformable Large Kernel Attention (D-LKA Attention), a streamlined attention mechanism employing large convolution kernels to fully appreciate volumetric context. This mechanism operates within a receptive field akin to self-attention while sidestepping the computational overhead. Additionally, our proposed attention mechanism benefits from deformable convolutions to flexibly warp the sampling grid, enabling the model to adapt appropriately to diverse data patterns. We designed both 2D and 3D adaptations of the D-LKA Attention, with the latter excelling in cross-depth data understanding. Together, these components shape our novel hierarchical Vision Transformer architecture, the D-LKA Net. Evaluations of our model against leading methods on popular medical segmentation datasets (Synapse, NIH Pancreas, and Skin lesion) demonstrate its superior performance.

Generic Object Tracking (GOT) is the problem of tracking target objects, specified by bounding boxes in the first frame of a video. While the task has received much attention in the last decades, researchers have almost exclusively focused on the single object setting. However multi-object GOT poses its own challenges and is more attractive in real-world applications. We attribute the lack of research interest into this problem to the absence of suitable benchmarks. In this work, we introduce a new large-scale GOT benchmark, LaGOT, containing multiple annotated target objects per sequence. Our benchmark allows users to tackle key remaining challenges in GOT, aiming to increase robustness and reduce computation through joint tracking of multiple objects simultaneously. In addition, we propose a transformer-based GOT tracker baseline capable of joint processing of multiple objects through shared computation. Our approach achieves a 4x faster run-time in case of 10 concurrent objects compared to tracking each object independently and outperforms existing single object trackers on our new benchmark. In addition, our approach achieves highly competitive results on single-object GOT datasets, setting a new state of the art on TrackingNet with a success rate AUC of 84.4%. Our benchmark, code, and trained models will be made publicly available.

Multispectral (MS) imaging systems have a wide range of applications for computer vision and computational photography tasks, but do not yet enjoy widespread adoption due to their prohibitive costs. Recently, advances in the design and fabrication of photonic metamaterials have enabled the development of MS sensors suitable for integration into consumer grade mobile devices. Augmenting existing RGB cameras and their processing algorithms with richer spectral information has the potential to be utilized in many steps of the image processing pipeline, but diverse real world datasets suitable for conducting such research are not freely available. We introduce Beyond RGB, a real-world dataset comprising thousands of multispectral and RGB images in diverse real world and lab conditions that is suitable for the development and evaluation of algorithms utilizing multispectral and RGB data. All the scenes in our dataset include a colorimetric reference and a measurement of the spectrum of the scene illuminant. Additionally, we demonstrate the practical use of our dataset through the introduction of a novel illuminant spectral estimation (ISE) algorithm. Our algorithm surpasses the current state-of-the-art (SoTA) by up to 58.6% on established benchmarks and sets a baseline performance on our own dataset.

Pedestrian detection is a fundamental task for many downstream applications. Visible and thermal images, as the two most important data types, are usually used to detect pedestrians under various environmental conditions. Many state-of-the-art works have been proposed to use two-stream (i.e., two-branch) architectures to combine visible and thermal information to improve detection performance. However, conventional visible-thermal fusion-based methods have no ability to obtain useful information from the visible branch under poor visibility conditions. The visible branch could even sometimes bring noise into the combined features. In this paper, we present a novel thermal and visible fusion architecture for pedestrian detection. Instead of simply using two branches to separately extract thermal and visible features and then fusing them, we introduce a hallucination branch to learn the mapping from thermal to visible domain, forming a three-branch feature extraction module. We then adaptively fuse feature maps from all the three branches (i.e., thermal, visible, and hallucination). With this new integrated hallucination branch, our network can still get relatively good visible feature maps under challenging low visibility conditions, thus boosting the overall detection performance. Finally, we experimentally demonstrate the superiority of the proposed architecture over conventional fusion methods.

This paper highlights the need to bring document classification benchmarking closer to real-world applications, both in the nature of data tested (X: multi-channel, multi-paged, multi-industry; Y: class distributions and label set variety) and in classification tasks considered (f: multi-page document, page stream, and document bundle classification, ...). We identify the lack of public multi-page document classification datasets, formalize different classification tasks arising in application scenarios, and motivate the value of targeting efficient multi-page document representations. An experimental study on proposed multi-page document classification datasets demonstrates that current benchmarks have become irrelevant and need to be updated to evaluate complete documents, as they naturally occur in practice. This reality check also calls for more mature evaluation methodologies, covering calibration evaluation, inference complexity (time-memory), and a range of realistic distribution shifts (e.g., born-digital vs. scanning noise, shifting page order). Our study ends on a hopeful note by recommending concrete avenues for future improvements.

Despite their impressive predictive performance in various computer vision tasks, deep neural networks (DNNs) tend to make overly confident predictions, which hinders their widespread use in safety-critical applications. While there have been recent attempts to calibrate DNNs, most of these efforts have primarily been focused on classification tasks, thus neglecting DNN-based object detectors. Although several recent works addressed calibration for object detection and proposed differentiable penalties, none of them are consistent estimators of established concepts in calibration. In this work, we tackle the challenge of defining and estimating calibration error specifically for this task. In particular, we adapt the definition of classification calibration error to handle the nuances associated with object detection, and predictions in structured output spaces more generally. Furthermore, we propose a consistent and differentiable estimator of the detection calibration error, utilizing kernel density estimation. Our experiments demonstrate the effectiveness of our estimator against competing train-time and post-hoc calibration methods, while maintaining similar detection performance.

Active Learning (AL) is a human-in-the-loop framework to interactively and adaptively label data instances, thereby enabling significant gains in model performance compared to random sampling. AL approaches function by selecting the hardest instances to label, often relying on notions of diversity and uncertainty. However, we believe that these current paradigms of AL do not leverage the full potential of human interaction granted by automated label suggestions. Indeed, we show that for many classification tasks and datasets, most people verifying if an automatically suggested label is correct take 3x to 4x less time than they do changing an incorrect suggestion to the correct label (or labeling from scratch without any suggestion). Utilizing this result, we propose CLARIFIER (aCtive LeARnIng From tIEred haRdness), an Interactive Learning framework that admits more effective use of human interaction by leveraging the reduced cost of verification. By targeting the hard (uncertain) instances with existing AL methods, the intermediate instances with a novel label suggestion scheme using submodular mutual information functions on a per-class basis, and the easy (confident) instances with highest-confidence auto-labeling, CLARIFIER can improve over the performance of existing AL approaches on multiple datasets -- particularly on those that have a large number of classes -- by almost 1.5x to 2x in terms of relative labeling cost.

Single image super-resolution (SISR) for reconstructing from a low-resolution (LR) input image its corresponding high-resolution (HR) output is a widely-studied research problem in the field of multimedia applications and computer vision. Despite the magic leap brought by recent development of deep neural networks for SISR, such problem is still considered to be quite challenging and non-scalable for the real-world data due to its ill-posed nature, where the degradations happened to the input LR images are usually complex and even unknown (in which the degradations in the test data could be unseen or different from the ones shown in the training dataset). To this end, two branches of SISR methods have emerged: blind super-resolution (blind-SR) and arbitrary-scale super-resolution (ASSR), where the former aims to reconstruct SR images under the unknown degradations, while the latter improves the scalability via learning to handle arbitrary up-sampling ratios. In this paper, we propose a holistic framework to take both blind-SR and ASSR tasks (accordingly named as arbitrary-scale blind-SR) into consideration with two main designs: 1) learning dual degradation representations where the implicit and explicit representations of degradation are sequentially extracted from the input LR image, and 2) modeling both upsampling (i.e. LR to HR) and downsampling (i.e. HR to LR) processes at the same time, where they utilize the implicit and explicit degradation representations respectively, in order to enable the cycle-consistency objective and further improve the training. We conduct extensive experiments on various datasets where the results well verify the effectiveness of our proposed framework in handling complex degradations as well as its superiority with respect to several state-of-the-art baselines.

When faced with an out-of-distribution (OOD) question or image, visual question answering (VQA) systems may provide unreliable answers. If relied on by real users or secondary systems, these failures may range from annoying to potentially endangering. Detecting OOD samples in single-modality settings is well-studied; however, limited attention has been paid to vision-and-language settings. In this work, we examine the question of OOD detection in the multimodal VQA task and benchmark a suite of approaches to identify OOD image-question pairs. In our experiments, we leverage popular VQA datasets to benchmark detection performance across a range of difficulties. We also produce composite datasets to examine impacts of individual modalities and of image-question agreement. Our results show that answer confidence alone is often a poor signal and that methods based on image-based question generation or examining model attention can lead to significantly better results. We find detecting ungrounded image-question pairs and small shifts in image distribution remain challenging.

We present a novel framework for generating adversarial benchmarks to evaluate the robustness of image classification models. The RLAB framework allows users to customize the types of distortions to be optimally applied to images, which helps address the specific distortions relevant to their deployment. The benchmark can generate datasets at various distortion levels to assess the robustness of different image classifiers. Our results show that the adversarial samples generated by our framework with any of the image classification models, like ResNet-50, Inception-V3, and VGG-16, are effective and transferable to other models causing them to fail. These failures happen even when these models are adversarially retrained using state-of-the-art techniques, demonstrating the generalizability of our adversarial samples. Our framework also allows the creation of adversarial samples for non-ground truth classes at different levels of intensity, enabling tunable benchmarks for the evaluation of false positives. We achieve competitive performance in terms of net L_2 distortion compared to state-of-the-art benchmark techniques on CIFAR-10 and ImageNet; however, we demonstrate our framework achieves such results with simple distortions like Gaussian noise without introducing unnatural artifacts or color bleeds. This is made possible by a model-based reinforcement learning (RL) agent and a technique that reduces a deep tree search of the image for model sensitivity to perturbations, to a one-level analysis and action. The flexibility of choosing distortions and setting classification probability thresholds for multiple classes makes our framework suitable for algorithmic audits.

Fully Test-Time Adaptation (TTA), which aims at adapting models to data drifts, has recently attracted wide interest. Numerous tricks and techniques have been proposed to ensure robust learning on arbitrary streams of unlabeled data. However, assessing the true impact of each individual technique and obtaining a fair comparison still constitutes a significant challenge. To help consolidate the community's knowledge, we present a categorization of selected orthogonal TTA techniques, including small batch normalization, stream rebalancing, reliable sample selection, and network confidence calibration. We meticulously dissect the effect of each approach on different scenarios of interest. Through our analysis, we shed light on trade-offs induced by those techniques between accuracy, the computational power required, and model complexity. We also uncover the synergy that arises when combining techniques and are able to establish new state-of-the-art results.

Learning-based methods have dominated the 3D human pose estimation (HPE) tasks with significantly better performance in most benchmarks than traditional optimization-based methods. Nonetheless, 3D HPE in the wild is still the biggest challenge of learning-based models, whether with 2D-3D lifting, image-to-3D, or diffusion-based methods, since the trained networks implicitly learn camera intrinsic parameters and domain-based 3D human pose distributions and estimate poses by statistical average. On the other hand, the optimization-based methods estimate results case-by-case, which can predict more diverse and sophisticated human poses in the wild. By combining the advantages of optimization-based and learning-based methods, we propose the Zero-shot Diffusion-based Optimization (ZeDO) pipeline for 3D HPE to solve the problem of cross-domain and in-the-wild 3D HPE. Our multi-hypothesis ZeDO achieves state-of-the-art (SOTA) performance on Human3.6M as minMPJPE 51.4mm without training with any 2D-3D or image-3D pairs. Moreover, our single-hypothesis ZeDO achieves SOTA performance on 3DPW dataset with PA-MPJPE 42.6mm on cross-dataset evaluation, which even outperforms learning-based methods trained on 3DPW.

In smartphones and compact cameras, the Image Signal Processor (ISP) transforms the RAW sensor image into a human-readable sRGB image. Most popular super-resolution methods depart from a sRGB image and upscale it further, improving its quality. However, modeling the degradations in the sRGB domain is complicated because of the non-linear ISP transformations. Despite this known issue, only a few methods work directly with RAW images and tackle real-world sensor degradations. We tackle blind image super-resolution in the RAW domain. We design a realistic degradation pipeline tailored specifically for training models with raw sensor data. Our approach considers sensor noise, defocus, exposure, and other common issues. Our BSRAW models trained with our pipeline can upscale real-scene RAW images and improve their quality. As part of this effort, we also present a new DSLM dataset and benchmark for this task.

Current knowledge distillation approaches in semantic segmentation tend to adopt a holistic approach that treats all spatial locations equally. However, for dense prediction, students' predictions on edge regions are highly uncertain due to contextual information leakage, requiring higher spatial sensitivity knowledge than the body regions. To address this challenge, this paper proposes a novel approach called boundary-privileged knowledge distillation (BPKD). it distils the knowledge from the teacher model's body and edges separately to the compact student model. Specifically, we employ two distinct loss functions: (i) edge loss, which aims to distinguish between ambiguous classes at the pixel level in edge regions; (ii) body loss, which utilizes shape constraints and selectively attends to the inner-semantic regions. Our experiments demonstrate that the proposed BPKD method provides extensive refinements and aggregation for edge and body regions. Additionally, the method achieves state-of-the-art distillation performance for semantic segmentation on three popular benchmark datasets, highlighting its effectiveness and generalization ability. BPKD shows consistent improvements across a diverse array of lightweight segmentation structures, including both CNNs and transformers, underscoring its architecture-agnostic adaptability.

In autonomous driving applications, there is a strong preference for modeling the world in Bird's-Eye View (BEV), as it leads to improved accuracy and performance. BEV features are widely used in perception tasks since they allow fusing information from multiple views in an efficient manner. However, BEV features generated from camera images are prone to be imprecise due to the difficulty of estimating depth in the perspective view. Improper placement of BEV features limits the accuracy of downstream tasks. We introduce a method for incorporating map information to improve perspective depth estimation from 2D camera images and thereby producing geometrically- and semantically-robust BEV features. We show that augmenting the camera images with the BEV map and map-to-camera projections can compensate for the depth uncertainty. Experiments on the nuScenes dataset demonstrate that our method outperforms previous approaches using only camera images in segmentation and detection tasks.

Local feature detection is a key ingredient of many image processing and computer vision applications, such as visual odometry and localization. Most existing algorithms focus on feature detection from a sharp image. They would thus have degraded performance once the image is blurred, which could happen easily under low-lighting conditions. To address this issue, we propose a simple yet both efficient and effective keypoint detection method that is able to accurately localize the salient keypoints in a blurred image. Our method takes advantages of a novel multi-layer perceptron (MLP) based architecture that significantly improve the detection repeatability for a blurred image. The network is also light-weight and able to run in real-time, which enables its deployment for time-constrained applications. Extensive experimental results demonstrate that our detector is able to improve the detection repeatability with blurred images, while keeping comparable performance as existing state-of-the-art detectors for sharp images. The code and trained weights are publicly available at github.com/ericzzj1989/BALF.