WACV 2023 论文列表

Unsupervised optical flow estimation relies on the assumption that pixels characterizing the same observed object should exhibit a stable appearance across video frames. With this assumption, the long-standing principle behind flow estimation consists in optimizing a photometric loss that maximizes the similarity between paired pixels in successive frames. However, these frames could be subject to strong brightness changes due to the radiometric properties of scenes as well as their viewing conditions. In this paper, we present BrightFlow, a new method to train any optical flow estimation network in an unsupervised manner. It consists in training two networks that jointly estimate optical flow and brightness changes. These changes are then compensated in the photometric loss so that reconstruction errors due to shadows or reflections will not affect negatively the training. As this compensation mechanism is only used at training stage, our method does not impact the number of parameters or the complexity at inference. Extensive experiments conducted on standard datasets and optical flow architectures show a consistent gain of our method. Source code is available at https://github.com/CEA-LIST/BrightFlow.

We present a new, simple yet effective approach to uplift video object detection. We observe that prior works operate on instance-level feature aggregation that imminently neglects the refined pixel-level representation, resulting in confusion among objects sharing similar appearance or motion characteristics. To address this limitation, we pro- pose BoxMask, which effectively learns discriminative representations by incorporating class-aware pixel-level information. We simply consider bounding box-level annotations as a coarse mask for each object to supervise our method. The proposed module can be effortlessly integrated into any region-based detector to boost detection. Extensive experiments on ImageNet VID and EPIC KITCHENS datasets demonstrate consistent and significant improvement when we plug our BoxMask module into numerous recent state-of-the-art methods. The code will be available at https://github.com/khurramHashmi/BoxMask.

Many state-of-the-art noisy-label learning methods rely on learning mechanisms that estimate the samples' clean labels during training and discard their original noisy labels. However, this approach prevents the learning of the relationship between images, noisy labels and clean labels, which has been shown to be useful when dealing with instance-dependent label noise problems. Furthermore, methods that do aim to learn this relationship require cleanly annotated subsets of data, as well as distillation or multi-faceted models for training. In this paper, we propose a new training algorithm that relies on a simple model to learn the relationship between clean and noisy labels without the need for a cleanly labelled subset of data. Our algorithm follows a 3-stage process, namely: 1) self-supervised pretraining followed by an early-stopping training of the classifier to confidently predict clean labels for a subset of the training set; 2) use the clean set from stage (1) to bootstrap the relationship between images, noisy labels and clean labels, which we exploit for effective relabelling of the remaining training set using semi-supervised learning; and 3) supervised training of the classifier with all relabelled samples from stage (2). By learning this relationship, we achieve state-of-the-art performance in asymmetric and instance-dependent label noise problems. Code is available at https://github.com/btsmart/bootstrapping-label-noise

The explosive increase in vision transformers studies has shown remarkable progress in vision tasks such as image classification and detection. However, in instance-level image retrieval, transformers have not yet shown good performance compared to convolutional networks. We propose a number of improvements that make transformers outperform the state of the art for the first time. (1) We show that a hybrid architecture is more effective than plain transformers, by a large margin. (2) We introduce two branches collecting global (classification token) and local (patch tokens) information, from which we form a global image epresentation. (3) In each branch, we collect multi-layer features from the transformer encoder, corresponding to skip connections across distant layers. (4) We enhance locality of interactions at the deeper layers of the encoder, which is the relative weakness of vision transformers. We train our model on all commonly used training sets and, for the first time, we make fair comparisons separately per training set. In all cases, we outperform previous models based on global representation.

In video compression, coding efficiency is improved by reusing pixels from previously decoded frames via motion and residual compensation. We define two levels of hierarchical redundancy in video frames: 1) first-order: redundancy in pixel space, i.e, similarities in pixel values across neighboring frames, which is effectively captured using motion and residual compensation, 2) second-order: redundancy in motion and residual maps due to smooth motion in natural videos. While most of the existing neural video coding literature addresses first-order redundancy, we tackle the problem of capturing second-order redundancy in neural video codecs via predictors. We introduce generic motion and residual predictors that learn to extrapolate from previously decoded data. These predictors are lightweight, and can be employed with most neural video codecs in order to improve their rate-distortion performance. Moreover, while RGB is the dominant colorspace in neural video coding literature, we introduce general modifications for neural video codecs to embrace the YUV420 colorspace and report YUV420 results. Our experiments show that using our predictors with a well-known neural video codec leads to 38% and 34% bitrate saving in RGB and YUV420 colorspaces measured on the UVG dataset.

Occluded person re-identification (ReID) is a person retrieval task which aims at matching occluded person images with holistic ones. For addressing occluded ReID, part-based methods have been shown beneficial as they offer fine-grained information and are well suited to represent partially visible human bodies. However, training a part-based model is a challenging task for two reasons. Firstly, individual body part appearance is not as discriminative as global appearance (two distinct IDs might have the same local appearance), this means standard ReID training objectives using identity labels are not adapted to local feature learning. Secondly, ReID datasets are not provided with human topographical annotations. In this work, we propose BPBreID, a body part-based ReID model for solving the above issues. We first design two modules for predicting body part attention maps and producing body part-based features of the ReID target. We then propose GiLt, a novel training scheme for learning part-based representations that is robust to occlusions and non-discriminative local appearance. Extensive experiments on popular holistic and occluded datasets show the effectiveness of our proposed method, which outperforms state-of-the-art methods by 0.7% mAP and 5.6% rank-1 accuracy on the challenging Occluded-Duke dataset. Our code is available at https://github.com/VlSomers/bpbreid.

Audio denoising has been explored for decades using both traditional and deep learning-based methods. However, these methods are still limited to either manually added artificial noise or lower denoised audio quality. To overcome these challenges, we collect a large-scale natural noise bird sound dataset. We are the first to transfer the audio denoising problem into an image segmentation problem and propose a deep visual audio denoising (DVAD) model. With a total of 14,120 audio images, we develop an audio ImageMask tool and propose to use a few-shot generalization strategy to label these images. Extensive experimental results demonstrate that the proposed model achieves state-of-the-art performance. We also show that our method can be easily generalized to speech denoising, audio separation, audio enhancement, and noise estimation.

Anomaly detection in video surveillance aims to detect anomalous frames whose properties significantly differ from normal patterns. Anomalies in videos can occur in both spatial appearance and temporal motion, making unsupervised video anomaly detection challenging. To tackle this problem, we investigate forward and backward motion continuity between adjacent frames and propose a new video anomaly detection paradigm based on bi-directional frame interpolation. The proposed framework consists of an optical flow estimation network and an interpolation network jointly optimized end-to-end to synthesize a middle frame from its nearest two frames. We further introduce a novel dynamic memory mechanism to balance memory sparsity and normality representation diversity, which attenuates abnormal features in frame interpolation without affecting normal prototypes. In inference, interpolation error and dynamic memory error are fused as anomaly scores. The proposed bi-directional interpolation design improves normal frame synthesis, lowering the false alarm rate of anomaly appearance; meanwhile, the implicit "regular" motion constraint in our optical flow estimation and the novel dynamic memory mechanism play blocking roles in interpolating abnormal frames, increasing the system's sensitivity to anomalies. Extensive experiments on public benchmarks demonstrate the superiority of the proposed framework over prior arts.

Comprehensive 3D scene understanding, both geometrically and semantically, is important for real-world applications such as robot perception. Most of the existing work has focused on developing data-driven discriminative models for scene understanding. This paper provides a new approach to scene understanding, from a synthesis model perspective, by leveraging the recent progress on implicit scene representation and neural rendering. Building upon the great success of Neural Radiance Fields (NeRFs), we introduce Scene-Property Synthesis with NeRF (SS-NeRF) that is able to not only render photo-realistic RGB images from novel viewpoints, but also render various accurate scene properties (e.g., appearance, geometry, and semantics). By doing so, we facilitate addressing a variety of scene understanding tasks under a unified framework, including semantic segmentation, surface normal estimation, reshading, keypoint detection, and edge detection. Our SS-NeRF framework can be a powerful tool for bridging generative learning and discriminative learning, and thus be beneficial to the investigation of a wide range of interesting problems, such as studying task relationships within a synthesis paradigm, transferring knowledge to novel tasks, facilitating downstream discriminative tasks as ways of data augmentation, and serving as auto-labeller for data creation. Our code is available at https://github.com/zsh2000/SS-NeRF.

Instance-level object re-identification is a fundamental computer vision task, with applications from image retrieval to intelligent monitoring and fraud detection. In this work, we propose the novel task of damaged object re-identification, which aims at distinguishing changes in visual appearance due to deformations or missing parts from subtle intra-class variations. To explore this task, we leverage the power of computer-generated imagery to create, in a semi-automatic fashion, high-quality synthetic images of the same bike before and after a damage occurs. The resulting dataset, Bent & Broken Bicycles (BBBicycles), contains 39,200 images and 2,800 unique bike instances spanning 20 different bike models. As a baseline for this task, we propose TransReI3D, a multi-task, transformer-based deep network unifying damage detection (framed as a multi-label classification task) with object re-identification.

This work introduces a simulator-based benchmark for visual localization in the autonomous navigation context. The dynamic benchmark enables investigation of how variables such as the time of day, weather, and camera perspective affect the navigation performance of autonomous agents that utilize visual localization for closed-loop control. The experimental part of the paper studies the effects of four such variables by evaluating state-of-the-art visual localization methods as part of the motion planning module of an autonomous navigation stack. The results show major variation in the suitability of the different methods for vision-based navigation. To the authors' best knowledge, the proposed benchmark is the first to study modern visual localization methods as part of a complete navigation stack. We make the benchmark available at https://github.com/lasuomela/carla_vloc_benchmark.

Visual question answering is a vision-and-language multimodal task, that aims at predicting answers given samples from the question and image modalities. Most recentmethods focus on learning a good joint embedding space ofimages and questions, either by improving the interactionbetween these two modalities, or by making it a more discriminant space. However, how informative this joint space is, has not been well explored. In this paper, we propose a novel regularization for VQA models, Constrained Optimization using Barlow's theory (COB), that improves the information content of the joint space by minimizing the redundancy. It reduces the correlation between the learned feature components and thereby disentangles semantic concepts. Our model also aligns the joint space with the answer embedding space, where we consider the answer and image+question as two different 'views' of what in essence is the same semantic information. We propose a constrained optimization policy to balance the categorical and redundancy minimization forces. When built on the state-of-the-art GGE model, the resulting model improves VQA accuracy by 1.4% and 4% on the VQA-CP v2 and VQA v2 datasets respectively. The model also exhibits better interpretability. Code is made available: https://github.com/abskjha/Barlow-constrained-VQA

The requirement for large labeled datasets is one of the limiting factors for training accurate deep neural networks. Unsupervised domain adaptation tackles this problem of limited training data by transferring knowledge from one domain, which has many labeled data, to a different domain for which little to no labeled data is available. One common approach is to learn domain-invariant features for example with an adversarial approach. Previous methods often train the domain classifier and label classifier network separately, where both classification networks have little interaction with each other. In this paper, we introduce a classifier-based backprop-induced weighting of the feature space. This approach has two main advantages. Firstly, it lets the domain classifier focus on features that are important for the classification, and, secondly, it couples the classification and adversarial branch more closely. Furthermore, we introduce an iterative label distribution alignment method, that employs results of previous runs to approximate a class-balanced dataloader. We conduct experiments and ablation studies on three benchmarks Office-31, OfficeHome, and DomainNet to show the effectiveness of our proposed algorithm.

This paper tackles the problem of human motion prediction, consisting in forecasting future body poses from historically observed sequences. State-of-the-art approaches provide good results, however, they rely on deep learning architectures of arbitrary complexity, such as Recurrent Neural Networks(RNN), Transformers or Graph Convolutional Networks(GCN), typically requiring multiple training stages and more than 2 million parameters. In this paper, we show that, after combining with a series of standard practices, such as applying Discrete Cosine Transform (DCT), predicting residual displacement of joints and optimizing velocity as an auxiliary loss, a light-weight network based on multi-layer perceptrons (MLPs) with only 0.14 million parameters can surpass the state-of-the-art performance. An exhaustive evaluation on the Human3.6M, AMASS, and 3DPW datasets shows that our method, named siMLPe, consistently outperforms all other approaches. We hope that our simple method could serve as a strong baseline for the community and allow re-thinking of the human motion prediction problem. The code is publicly available at https://github.com/dulucas/siMLPe.

Multiple existing benchmarks involve tracking and segmenting objects in video e.g., Video Object Segmentation (VOS) and Multi-Object Tracking and Segmentation (MOTS), but there is little interaction between them due to the use of disparate benchmark datasets and metrics (e.g. \JnF, mAP, sMOTSA). As a result, published works usually target a particular benchmark, and are not easily comparable to each another. We believe that the development of generalized methods that can tackle multiple tasks requires greater cohesion among these research sub-communities. In this paper, we aim to facilitate this by proposing BURST, a dataset which contains thousands of diverse videos with high-quality object masks, and an associated benchmark with six tasks involving object tracking and segmentation in video. All tasks are evaluated using the same data and comparable metrics, which enables researchers to consider them in unison, and hence, more effectively pool knowledge from different methods across different tasks. Additionally, we demonstrate several baselines for all tasks and show that approaches for one task can be applied to another with a quantifiable and explainable performance difference. Dataset annotations are available at: https://github.com/Ali2500/BURST-benchmark.

Semantic segmentation in bird's eye view (BEV) is an important task for autonomous driving. Though this task has attracted a large amount of research efforts, it is still challenging to flexibly cope with arbitrary (single or multiple) camera sensors equipped on the autonomous vehicle. In this paper, we present BEVSegFormer, an effective transformer-based method for BEV semantic segmentation from arbitrary camera rigs. Specifically, our method first encodes image features from arbitrary cameras with a shared backbone. These image features are then enhanced by a deformable transformer-based encoder. Moreover, we introduce a BEV transformer decoder module to parse BEV semantic segmentation results. An efficient multi-camera deformable attention unit is designed to carry out the BEV-to-image view transformation. Finally, the queries are reshaped according the layout of grids in the BEV, and upsampled to produce the semantic segmentation result in a supervised manner. We evaluate the proposed algorithm on the public nuScenes dataset and a self-collected dataset. Experimental results show that our method achieves promising performance on BEV semantic segmentation from arbitrary camera rigs. We also demonstrate the effectiveness of each component via ablation study.

This paper considers the active recognition scenario, where the agent is empowered to intelligently acquire observations for better recognition. The agents usually compose two modules, i.e., the policy and the recognizer, to select actions and predict the category. While using ground-truth class labels to supervise the recognizer, the policy is typically updated with rewards determined by the current in-training recognizer, like whether achieving correct predictions. However, this joint learning process could lead to unintended solutions, like a collapsed policy that only visits views that the recognizer is already sufficiently trained to obtain rewards, which harms the generalization ability. We call this phenomenon lingering to depict the agent being reluctant to explore challenging views during training. Existing approaches to tackle the exploration-exploitation trade-off could be ineffective as they usually assume reliable feedback during exploration to update the estimate of rarely-visited states. This assumption is invalid here as the reward from the recognizer could be insufficiently trained. To this end, our approach integrates another adversarial policy to constantly disturb the recognition agent during training, forming a competing game to promote active explorations and avoid lingering. The reinforced adversary, rewarded when the recognition fails, contests the recognition agent by turning the camera to challenging observations. Extensive experiments across two datasets validate the effectiveness of the proposed approach regarding its recognition performances, learning efficiencies, and especially robustness in managing environmental noises.

The research area of domain adaptation investigates methods that enable the transfer of existing models across different domains, e.g., addressing environmental changes or the transfer from synthetic to real data. Especially unsupervised domain adaptation is beneficial because it does not require any labeled target domain data. Usually, existing methods are targeted at specific tasks and require access or even modifications to the source model and its parameters which is a major drawback when only a black-box model is available. Therefore, we propose a CycleGAN-based approach suitable for black-box source models to translate target domain data into the source domain on which the source model can operate. Inspired by multi-task learning, we extend CycleGAN with an additional auxiliary task that can be arbitrarily chosen to support the transfer of task-related information across domains without the need for having access to a differentiable source model or its parameters. In this work, we focus on the regression task of 2D human pose estimation and compare our results in four different domain adaptation settings to CycleGAN and RegDA, a state-of-the-art method for unsupervised domain adaptation for keypoint detection.

We present an automatic method for annotating images of indoor scenes with the CAD models of the objects by relying on RGB-D scans. Through a visual evaluation by 3D experts, we show that our method retrieves annotations that are at least as accurate as manual annotations, and can thus be used as ground truth without the burden of manually annotating 3D data. We do this using an analysis-by-synthesis approach, which compares renderings of the CAD models with the captured scene. We introduce a 'cloning procedure' that identifies objects that have the same geometry, to annotate these objects with the same CAD models. This allows us to obtain complete annotations for the ScanNet dataset and the recent ARKitScenes dataset. We will release these annotations publicly, as we believe they will be very useful for the computer vision community.