WACV 2024 论文列表

Although instance segmentation methods have improved considerably, the dominant paradigm is to rely on fully annotated training images, which are tedious to obtain. To alleviate this reliance, and boost results, semi-supervised approaches leverage unlabeled data as an additional training signal that limits overfitting to the labeled samples. In this context, we present novel design choices to significantly improve teacher-student distillation models. In particular, we (i) improve the distillation approach by introducing a novel "guided burn-in" stage, and (ii) evaluate different instance segmentation architectures, as well as backbone networks and pre-training strategies. Contrary to previous work which uses only supervised data for the burn-in period of the student model, we also use guidance of the teacher model to exploit unlabeled data in the burn-in period. Our improved distillation approach leads to substantial improvements over previous state-of-the-art results. For example, on the Cityscapes dataset we improve mask-AP from 23.7 to 33.9 when using labels for 10% of images, and on the COCO dataset we improve mask-AP from 18.3 to 34.1 when using labels for only 1% of the training data.

Despite advances in image recognition, recognizing novel categories in unlabeled data remains challenging for machine learning methods, even though humans can perform this task with ease. A recently developed setting to tackle this problem is Generalized Category Discovery (GCD), in which the task is to, given a labeled dataset, classify an unlabeled dataset, where the unlabeled dataset contains both known classes and novel classes that do not appear in the labeled data. Existing GCD methods mostly focus on learning strong image representations, on which they then apply a clustering algorithm such as k-means. Despite obtaining good performance, they do not fully exploit the potential of the learned features due to the simple nature of the clustering mechanism. To address this issue, we make use of the fact that local neighborhoods in self-supervised feature spaces are highly homogeneous. We leverage this observation to develop Guided Cluster Aggregation (GCA), a hierarchical approach that first groups the data into small clusters of high purity, then aggregates them into larger clusters. Experiments show that GCA outperforms semi-supervised k-means in most cases, especially in fine-grained classification tasks. Code available at https://github.com/J- L- O/guided-cluster-aggregation.

Coarse or weak labels can serve as a cost-effective solution to the problem of visual representation learning. When fine-grained labels are unavailable, weak labels can provide some form of supervisory signals to guide the representation learning process. Some examples of weak labels include image captions, visual attributes and coarse-grained object categories. In this work, we consider the semantic grouping relationship that exists within certain types of weak labels and propose a group-wise contrastive bottleneck module to leverage this relationship. The semantic group may contain labels that are related to a general concept, such as the colour or shape of objects. Using the group-wise bottleneck module, we disentangle the global image features into multiple group features and apply contrastive learning in a group-wise manner to maximize the similarity of positive pairs within each semantic group. The positive pairs are defined based on the similarity of the labels captured by each group. To learn a more robust representation, we introduce a reconstruction objective where an image feature is reconstructed back from the disentangled features, and this reconstruction is encouraged to be consistent with the feature obtained from a different augmented view of the same image. We empirically verify the efficacy of the proposed method on several datasets in the context of visual attribute learning, fair representation learning and hierarchical label learning. The experimental results indicate that our proposed method outperforms prior weakly-supervised methods and is flexible in adapting to different representation learning settings.

We present a novel coarser-to-finer approach for deep graphical image inpainting that utilizes GraphFill, a graph neural network-based deep learning framework, and a lightweight generative baseline network. We construct a pyramidal graph for the input-masked image by reducing it into superpixels, each representing a node in the graph. The proposed pyramidal approach facilitates the transfer of global context from coarser to finer pyramid levels, enabling GraphFill to estimate plausible information for unknown node values in the graph. The estimated information is used to fill in the masked region, which a Refine Network then refines. Furthermore, we propose a resolution-robust pyramidal graph construction method, allowing for efficient inpainting of high-resolution images with relatively fewer computations. Our proposed network, trained on Places and CelebA-HQ datasets, demonstrates competitive performance compared to existing methods while using fewer learning parameters. We conduct thorough ablation studies to evaluate the effectiveness of each component in the GraphFill Network for improved performance. Our proposed lightweight model for image inpainting is efficient in real-world scenarios, as it can be easily deployed on mobile devices with limited resources.

Anticipating traffic accidents early using dashcam videos is an important task for ensuring road safety and building reliable intelligent autonomous vehicles. However, factors like high traffic on the roads, different types of accidents, limited angles of vision, etc. make this task very challenging. Using the early frames, a lot of existing methods predict a large number of false positives which poses a huge risk for all vehicles on the road. In this paper, we propose a novel end-to-end learning, nested graph-based framework named Graph(Graph) for early accident anticipation. It uses interactions between the objects in the same as well as the neighboring frames along with the global features to make precise predictions as early as possible. This way it is able to embed the local as well as global temporal information into the extracted features. Graph(Graph) outperforms state-of-the-art methods on different datasets by a large margin demonstrating its effectiveness. With empirical evidence, we highlight the importance of each component in Graph(Graph) and show their effect on the final performance. Our code is available at https://github.com/thakurnupur/Graph-Graph.

Automating Information Extraction (IE) from handwritten documents is a challenging task due to the wide variety of handwriting styles, the presence of noise, and the lack of labeled data. In this work, we propose an end-to-end encoder-decoder model, that incorporates transformers and Graph Convolutional Networks (GCN), to jointly perform Handwritten Text Recognition (HTR) and Named Entity Recognition (NER). The proposed architecture is mainly composed of two parts: a Sparse Graph Transformer Encoder (SGTE), to capture efficient representations of input text images while controlling the propagation of information through the model. The SGTE is followed by a transformer decoder enhanced with a GCN that combines the outputs of the last SGTE layer and the Multi-Head Attention (MHA) block to reinforce the alignment of visual features to characters and Named Entity (NE) tags, resulting in more robust learned representations. The proposed model shows promising results and achieves state-of-the-art performance on the IAM dataset, and in the ICDAR 2017 Information Extraction competition using the Esposalles database.

Vision Transformers (ViTs) have recently become the state-of-the-art across many computer vision tasks. In contrast to convolutional networks (CNNs), ViTs enable global information sharing even within shallow layers of a network, i.e., among high-resolution features. However, this perk was later overlooked with the success of pyramid architectures such as Swin Transformer, which show better performance-complexity trade-offs. In this paper, we present a simple and efficient add-on component (termed GrafT) that considers global dependencies and multi-scale information throughout the network, in both high- and low-resolution features alike. It has the flexibility of branching out at arbitrary depths and shares most of the parameters and computations of the backbone. GrafT shows consistent gains over various well-known models which includes both hybrid and pure Transformer types, both homogeneous and pyramid structures, and various self-attention methods. In particular, it largely benefits mobile-size models by providing high-level semantics. On the ImageNet-1k dataset, GrafT delivers +3.9%, +1.4%, and +1.9% top-1 accuracy improvement to DeiT-T, Swin-T, and MobileViT-XXS, respectively. The code and models are at https://github.com/jongwoopark7978/Grafting-Vision-Transformer.

Unsupervised domain adaptation (UDA) tries to overcome the need of a large labeled dataset by transferring knowledge from a source dataset, with lots of labeled data, to a target dataset, that has no labeled data. Since there are no labels in the target domain, early misalignment might propagate into the later stages and lead to an error build-up. In order to overcome this problem, we propose a gradual source domain expansion (GSDE) algorithm. GSDE trains the UDA task several times from scratch, but each time expands the source dataset with target data. In particular, the highest scoring target data of the previous run are employed as pseudo-source samples with their respective pseudo-label. Using this strategy, the pseudo source samples induce knowledge extracted from the previous run directly from the start of the new training. This helps align the two domains better especially in the early training epochs. In this study, we first introduce a strong baseline network and apply our GSDE strategy to it. We conduct experiments and ablation studies on three benchmarks (Office-31, OfficeHome, and DomainNet) and outperform state-of-the-art methods. We further show that the proposed GSDE strategy can improve the accuracy of a variety of different state-of-the-art UDA approaches.

Deep learning models have demonstrated remarkable success in object detection, yet their complexity and computational intensity pose a barrier to deploying them in real-world applications (e.g., self-driving perception). Knowledge Distillation (KD) is an effective way to derive efficient models. However, only a small number of KD methods tackle object detection. Also, most of them focus on mimicking the plain features of the teacher model but rarely consider how the features contribute to the final detection. In this paper, we propose a novel approach for knowledge distillation in object detection, named Gradient-guided Knowledge Distillation (GKD). Our GKD uses gradient information to identify and assign more weights to features that significantly impact the detection loss, allowing the student to learn the most relevant features from the teacher. Furthermore, we present bounding-box-aware multi-grained feature imitation (BMFI) to further improve the KD performance. Experiments on the KITTI and COCO Traffic datasets demonstrate our method's efficacy in knowledge distillation for object detection. On one-stage and two-stage detectors, our GKD-BMFI leads to an average of 5.1% and 3.8% mAP improvement, respectively, beating various state-of-the-art KD methods. Our codes are available at: https://github.com/lanqz7766/GKD.

Federated Learning (FL) is used to learn machine learning models with data that is partitioned across multiple clients, including resource-constrained edge devices. It is therefore important to devise solutions that are efficient in terms of compute, communication, and energy consumption, while ensuring compliance with the FL framework's privacy requirements. Conventional approaches to these problems select a weighted subset of the training dataset, known as coreset, and learn by fitting models on it. Such coreset selection approaches are also known to be robust to data noise. However, these approaches rely on the overall statistics of the training data and are not easily extendable to the FL setup. In this paper, we propose an algorithm called Gradient based Coreset for Robust and Efficient Federated Learning (GCFL) that selects a coreset at each client, only every K communication rounds and derives updates only from it, assuming the availability of a small validation dataset at the server. We demonstrate that our coreset selection technique is highly effective in accounting for noise in clients' data. We conduct experiments using four real-world datasets and show that GCFL is (1) more compute and energy efficient than FL, (2) robust to various kinds of noise in both the feature space and labels, (3) preserves the privacy of the validation dataset, and (4) introduces a small communication overhead but achieves significant gains in performance, particularly in cases when the clients' data is noisy.

Despite the remarkable progress facilitated by learning-based stereo-matching algorithms, the performance in the ill-conditioned regions, such as the occluded regions, remains a bottleneck. Due to the limited receptive field, existing CNN-based methods struggle to handle these ill-conditioned regions effectively. To address this issue, this paper introduces a novel attention-based stereo-matching network called Global Occlusion-Aware Transformer (GOAT) to exploit long-range dependency and occlusion-awareness global context for disparity estimation. In the GOAT architecture, a parallel disparity and occlusion estimation module PDO is proposed to estimate the initial disparity map and the occlusion mask using a parallel attention mechanism. To further enhance the disparity estimates in the occluded regions, an occlusion-aware global aggregation module (OGA) is proposed. This module aims to refine the disparity in the occluded regions by leveraging restricted global correlation within the focus scope of the occluded areas. Extensive experiments were conducted on several public benchmark datasets including SceneFlow, KITTI 2015, and Middlebury. The results show that the proposed GOAT demonstrates outstanding performance among all benchmarks, particularly in the occluded regions.

Crowd image is arguably one of the most laborious data to annotate. In this paper, we devote to reduce the massive demand of densely labeled crowd data, and propose a novel weakly-supervised setting, in which we leverage the binary ranking of two images with highcontrast crowd counts as training guidance. To enable training under this new setting, we convert the crowd count regression problem to a ranking potential prediction problem. In particular, we tailor a Siamese Ranking Network that predicts the potential scores of two images indicating the ordering of the counts. Hence, the ultimate goal is to assign appropriate potentials for all the crowd images to ensure their orderings obey the ranking labels. On the other hand, potentials reveal the relative crowd sizes but cannot yield an exact crowd count. We resolve this problem by introducing "anchors" during the inference stage. Concretely, anchors are a few images with count labels used for referencing the corresponding counts from potential scores by a simple linear mapping function. We conduct extensive experiments to study various combinations of supervision, and we show that the proposed method outperforms existing weakly-supervised methods without additional labeling effort by a large margin.

The inclination of a spherical camera results in nonupright panoramic images. To carry out upright adjustment, traditional methods estimate camera inclination angles firstly, and then resample the image in terms of the estimated rotation to generate upright image. Since sampling an image is a time-consuming processing, a lookup table is usually used to achieve a high processing speed; however, the content of a lookup table depends on the rotational angles and needs extra memory to store also. In this paper we propose a new approach for panorama upright adjustment, which directly generates an upright panoramic image from an input nonupright one without rotation estimation and lookup tables as an intermediate processing. The proposed approach formulates panorama upright adjustment as a pixelwise image-to-image mapping problem, and the mapping is directly generated from an input nonupright panoramic image via an end-to-end neural network. As shown in the experiment of this paper, the proposed method results in a lightweight network, as less as 163MB, with high processing speed, as great as 9ms, for a 256x512 pixel panoramic image.

Generative models have demonstrated revolutionary success in various visual creation tasks, but in the meantime, they have been exposed to the threat of leaking private information of their training data. Several membership inference attacks (MIAs) have been proposed to exhibit the privacy vulnerability of generative models by classifying a query image as a training dataset member or nonmember. However, these attacks suffer from major limitations, such as requiring shadow models and white-box access, and either ignoring or only focusing on the unique property of diffusion models, which block their generalization to multiple generative models. In contrast, we propose the first generalized membership inference attack against a variety of generative models such as generative adversarial networks, [variational] autoencoders, implicit functions, and the emerging diffusion models. We leverage only generated distributions from target generators and auxiliary non-member datasets, therefore regarding target generators as black boxes and agnostic to their architectures or application scenarios. Experiments validate that all the generative models are vulnerable to our attack. For instance, our work achieves attack AUC > 0.99 against DDPM, DDIM, and FastDPM trained on CIFAR-10 and CelebA. And the attack against VQGAN, LDM (for the text-conditional generation), and LIIF achieves AUC > 0.90. As a result, we appeal to our community to be aware of such privacy leakage risks when designing and publishing generative models.

Diabetic retinopathy (DR) is caused by long-standing diabetes and is among the fifth leading cause for visual impairment. The prospects of early diagnosis and treatment could be helpful in curing the disease, however, the detection procedure is rather challenging and mostly tedious. Therefore, automated diabetic retinopathy classification using deep learning techniques has gained interest in the medical imaging community. Akin to several other real-world applications of deep learning, the typical assumption of i.i.d data is also violated in DR classification that relies on deep learning. Therefore, developing DR classification methods robust to unseen distributions is of great value. In this paper, we study the problem of generalizing a model to unseen distributions or domains (a.k.a domain generalization) in DR classification. To this end, we propose a simple and effective domain generalization (DG) approach that achieves self-distillation in vision transformers (ViT) via a novel prediction softening mechanism. This prediction softening is an adaptive convex combination of one-hot labels with the model's own knowledge. We perform extensive experiments on challenging open-source DR classification datasets under both multi-source and more challenging single-source DG settings with three different ViT backbones to establish the efficacy and applicability of our approach against competing methods. For the first time, we report the performance of several state-of-the-art domain generalization (DG) methods on open-source DR classification datasets after conducting thorough experiments. Finally, our method is also capable of delivering improved calibration performance than other methods, showing its suitability for safety-critical applications, including healthcare. We hope that our contributions would instigate more DG research across the medical imaging community.

When models, e.g., for semantic segmentation, are applied to images that are vastly different from training data, the performance will drop significantly. Domain adaptation methods try to overcome this issue, but need samples from the target domain. However, this might not always be feasible for various reasons and therefore domain generalization methods are useful as they do not require any target data. We present a new diffusion-based domain extension (DIDEX) method and employ a diffusion model to generate a pseudo-target domain with diverse text prompts. In contrast to existing methods, this allows to control the style and content of the generated images and to introduce a high diversity. In a second step, we train a generalizing model by adapting towards this pseudo-target domain. We outperform previous approaches by a large margin across various datasets and architectures without using any real data. For the generalization from GTA5, we improve state-of-the-art mIoU performance by 3.8% absolute on average and for SYNTHIA by 11.8% absolute, marking a big step for the generalization performance on these benchmarks. Code is available at https://github.com/JNiemeijer/DIDEX

Eye tracking research is important in computer vision because it can help us understand how humans interact with the visual world. Specifically for high-risk applications, such as in medical imaging, eye tracking can help us to comprehend how radiologists and other medical professionals search, analyze, and interpret images for diagnostic and clinical purposes. Hence, the application of eye tracking techniques in disease classification has become increasingly popular in recent years. Contemporary works usually transform gaze information collected by eye tracking devices into visual attention maps (VAMs) to supervise the learning process. However, this is a time-consuming preprocessing step, which stops us from applying eye tracking to radiologists' daily work. To solve this problem, we propose a novel gaze-guided graph neural network (GNN), GazeGNN, to leverage raw eye-gaze data without being converted into VAMs. In GazeGNN, to directly integrate eye gaze into image classification, we create a unified representation graph that models both images and gaze pattern information. With this benefit, we develop a real-time, real-world, end-to-end disease classification algorithm for the first time in the literature. This achievement demonstrates the practicality and feasibility of integrating real-time eye tracking techniques into the daily work of radiologists. To our best knowledge, GazeGNN is the first work that adopts GNN to integrate image and eye-gaze data. Our experiments on the public chest X-ray dataset show that our proposed method exhibits the best classification performance compared to existing methods. The code is available.

Vision Transformers (ViTs) have revolutionized the field of computer vision, yet their deployments on resource-constrained devices remain challenging due to high computational demands. To expedite pre-trained ViTs, token pruning and token merging approaches have been developed, which aim at reducing the number of tokens involved in the computation. However, these methods still have some limitations, such as image information loss from pruned tokens and inefficiency in the token-matching process. In this paper, we introduce a novel Graph-based Token Propagation (GTP) method to resolve the challenge of balancing model efficiency and information preservation for efficient ViTs. Inspired by graph summarization algorithms, GTP meticulously propagates less significant tokens' information to spatially and semantically connected tokens that are of greater importance. Consequently, the remaining few tokens serve as a summarization of the entire token graph, allowing the method to reduce computational complexity while preserving essential information of eliminated tokens. Combined with an innovative token selection strategy, GTP can efficiently identify image tokens to be propagated. Extensive experiments have validated GTP's effectiveness, demonstrating both efficiency and performance improvements. Specifically, GTP decreases the computational complexity of both DeiT-S and DeiT-B by up to 26% with only a minimal 0.3% accuracy drop on ImageNet-1K without finetuning, and remarkably surpasses the state-of-the-art token merging method on various backbones at an even faster inference speed. The source code is available in the supplementary material.

Current Image-to-Image translation (I2I) frameworks rely heavily on reconstruction losses, where the output needs to match a given ground truth image. An adversarial loss is commonly utilized as a secondary loss term, mainly to add more realism to the output. Compared to unconditional GANs, I2I translation frameworks have more supervisory signals, but still their output shows more artifacts and does not reach the same level of realism achieved by unconditional GANs. We study the performance gap, in terms of photo-realism, between I2I translation and unconditional GAN frameworks. Based on our observations, we propose a modified architecture and training objective to address this realism gap. Our proposal relaxes the role of reconstruction losses, to act as regularizers instead of doing all the heavy lifting which is common in current I2I frameworks. Furthermore, our proposed formulation decouples the optimization of reconstruction and adversarial objectives and removes pixel-wise constraints on the final output. This allows for a set of stochastic but realistic variations of any target output image.