WACV 2023 Papers

Learning self-supervised video representation predominantly focuses on discriminating instances generated from simple data augmentation schemes. However, the learned representation often fails to generalize over unseen camera viewpoints. To this end, we propose ViewCLR, that learns self-supervised video representation invariant to camera viewpoint changes. We introduce a viewpoint-generator that can be considered as a learnable augmentation for any self-supervised pre-text tasks, to generate latent viewpoint representation of a video. ViewCLR maximizes the similarities between the representation of the latent viewpoint and that of the original viewpoint, enabling the learned video encoder to generalize over unseen camera viewpoints. Experiments on cross-view benchmark datasets including NTU RGB+D dataset show that ViewCLR stands as a state-of-the-art viewpoint invariant self-supervised method.

Different from most existing approaches that require trimap generation for each frame, we reformulate video object matting (VOM) by introducing improved semantic guidance propagation. The proposed approach can achieve a higher degree of temporal coherence between frames with only a single coarse mask as reference. In this paper, we adapt the hierarchical memory matching mechanism into the space-time baseline to build an efficient and robust framework for semantic guidance propagation and alpha prediction. To enhance the temporal smoothness, we also propose a cross-frame attention refinement (CFAR) module that can refine the feature representations across multiple adjacent frames (both historical and current frames) based on the spatio-temporal correlation among the cross-frame pixels. Extensive experiments demonstrate the effectiveness of hierarchical spatio-temporal semantic guidance and the cross-video-frame attention refinement module, and our model outperforms the state-of-the-art VOM methods. We also analyze the significance of different components in our model.

Denoising and demosaicing are two critical components of the image/video processing pipeline. While historically these two tasks have mainly been considered separately, current neural network approaches allow to obtain state-of-the-art results by treating them jointly. However, most existing research focuses in single image or burst joint denoising and demosaicing (JDD). Although related to burst JDD, video JDD deserves its own treatment. In this work we present an empirical exploration of different design aspects of video joint denoising and demosaicing using neural networks. We compare recurrent and non-recurrent approaches and explore aspects such as type of propagated information in recurrent networks, motion compensation, video stabilization, and network architecture. We found that recurrent networks with motion compensation achieve best results. Our work should serve as a strong baseline for future research in video JDD.

Detection of melanocytes serves as a critical prerequisite in assessing melanocytic growth patterns when diagnosing melanoma and its precursor lesions on skin biopsy specimens. However, this detection is challenging due to the visual similarity of melanocytes to other cells in routine Hematoxylin and Eosin (H&E) stained images, leading to the failure of current nuclei detection methods. Stains such as Sox10 can mark melanocytes, but they require an additional step and expense and thus are not regularly used in clinical practice. To address these limitations, we introduce VSGD-Net, a novel detection network that learns melanocyte identification through virtual staining from H&E to Sox10. The method takes only routine H&E images during inference, resulting in a promising approach to support pathologists in the diagnosis of melanoma. To the best of our knowledge, this is the first study that investigates the detection problem using image synthesis features between two distinct pathology stainings. Extensive experimental results show that our proposed model outperforms state-of-the-art nuclei detection methods.

There has been a growing interest in solving Visual Question Answering (VQA) tasks that require the model to reason beyond the content present in the image. In this work, we focus on questions that require commonsense reasoning. In contrast to previous methods which inject knowledge from static knowledge bases, we investigate the incorporation of contextualized knowledge using Commonsense Transformer (COMET), an existing knowledge model trained on human-curated knowledge bases. We propose a method to generate, select, and encode external commonsense knowledge alongside visual and textual cues in a new pre-trained Vision-Language-Commonsense transformer model, VLC-BERT. Through our evaluation on the knowledge-intensive OK-VQA and A-OKVQA datasets, we show that VLC-BERT is capable of outperforming existing models that utilize static knowledge bases. Furthermore, through a detailed analysis, we explain which questions benefit, and which don't, from contextualized commonsense knowledge from COMET.

For best performance, today's semantic segmentation methods use large and carefully labeled datasets, requiring expensive annotation budgets. In this work, we show that coarse annotation is a low-cost but highly effective alternative for training semantic segmentation models. Considering the urban scene segmentation scenario, we leverage cheap coarse annotations for real-world captured data, as well as synthetic data to train our model and show competitive performance compared with fully annotated real-world data. Specifically, we propose a coarse-to fine self-training framework that generates pseudo labels for unlabeled regions of the coarsely annotated data, using synthetic data to improve predictions around the boundaries between semantic classes, and using cross-domain data augmentation to increase diversity. Our extensive experimental results on Cityscapes and BDD100k datasets demonstrate that our method achieves a significantly better performance vs annotation cost tradeoff, yielding a comparable performance to fully annotated data with only a small fraction of the annotation budget. Also, when used as pretraining, our framework performs better compared to the standard fully supervised setting.

The state-of-the-art for monocular 3D human pose estimation in videos is dominated by the paradigm of 2D-to-3D pose uplifting. While the uplifting methods themselves are rather efficient, the true computational complexity depends on the per-frame 2D pose estimation. In this paper, we present a Transformer-based pose uplifting scheme that can operate on temporally sparse 2D pose sequences but still produce temporally dense 3D pose estimates. We show how masked token modeling can be utilized for temporal upsampling within Transformer blocks. This allows to decouple the sampling rate of input 2D poses and the target frame rate of the video and drastically decreases the total computational complexity. Additionally, we explore the option of pre-training on large motion capture archives, which has been largely neglected so far. We evaluate our method on two popular benchmark datasets: Human3.6M and MPI-INF-3DHP. With an MPJPE of 45.0 mm and 46.9 mm, respectively, our proposed method can compete with the state-of-the-art while reducing inference time by a factor of 12. This enables real-time throughput with variable consumer hardware in stationary and mobile applications. We release our code and models at https://github.com/goldbricklemon/uplift-upsample-3dhpe

Unsupervised video object segmentation aims to segment a target object in the video without a ground truth mask in the initial frame. This challenging task requires extracting features for the most salient common objects within a video sequence. This difficulty can be solved by using motion information such as optical flow, but using only the information between adjacent frames results in poor connectivity between distant frames and poor performance. To solve this problem, we propose a novel prototype memory network architecture. The proposed model effectively extracts the RGB and motion information by extracting superpixel-based component prototypes from the input RGB images and optical flow maps. In addition, the model scores the usefulness of the component prototypes in each frame based on a self-learning algorithm and adaptively stores the most useful prototypes in memory and discards obsolete prototypes. We use the prototypes in the memory bank to predict the next query frame's mask, which enhances the association between distant frames to help with accurate mask prediction. Our method is evaluated on three datasets, achieving state-of-the-art performance. We prove the effectiveness of the proposed model with various ablation studies.

We introduce a new architecture for unsupervised object-centric representation learning and multi-object detection and segmentation, which uses a translation-equivariant attention mechanism to predict the coordinates of the objects present in the scene and to associate a feature vector to each object. A transformer encoder handles occlusions and redundant detections, and a convolutional autoencoder is in charge of background reconstruction. We show that this architecture significantly outperforms the state of the art on complex synthetic benchmarks.

Over the last decade, online lecture videos have become increasingly popular and have experienced a meteoric rise during the pandemic. However, video-language research has primarily focused on instructional videos or movies, and tools to help students navigate the growing online lectures are lacking. Our first contribution is to facilitate research in the educational domain, by introducing AVLectures, a large-scale dataset consisting of 86 courses with over 2,350 lectures covering various STEM subjects. Each course contains video lectures, transcripts, OCR outputs for lecture frames, and optionally lecture notes, slides, assignments, and related educational content that can inspire a variety of tasks. Our second contribution is introducing video lecture segmentation that splits lectures into bite-sized topics that show promise in improving learner engagement. We formulate lecture segmentation as an unsupervised task that leverages visual, textual, and OCR cues from the lecture, while clip representations are fine-tuned on a pretext self-supervised task of matching the narration with the temporally aligned visual content. We use these representations to generate segments using a temporally consistent 1-nearest neighbor algorithm, TW-FINCH. We evaluate our method on 15 courses and compare it against various visual and textual baselines, outperforming all of them. Our comprehensive ablation studies also identify the key factors driving the success of our approach.

In this work, we address the problem of unsupervised moving object segmentation (MOS) in 4D LiDAR data recorded from a stationary sensor, where no ground truth annotations are involved. Deep learning-based state-of-the-art methods for LiDAR MOS strongly depend on annotated ground truth data, which is expensive to obtain and scarce in existence. To close this gap in the stationary setting, we propose a novel 4D LiDAR representation based on multivariate time series that relaxes the problem of unsupervised MOS to a time series clustering problem. More specifically, we propose modeling the change in occupancy of a voxel by a multivariate occupancy time series (MOTS), which captures spatio-temporal occupancy changes on the voxel level and its surrounding neighborhood. To perform unsupervised MOS, we train a neural network in a self-supervised manner to encode MOTS into voxel-level feature representations, which can be partitioned by a clustering algorithm into moving or stationary. Experiments on stationary scenes from the Raw KITTI dataset show that our fully unsupervised approach achieves performance that is comparable to that of supervised state-of-the-art approaches.

Deep image compression performs better than conventional codecs, such as JPEG, on natural images. However, deep image compression is learning-based and encounters a problem: the compression performance deteriorates significantly for out-of-domain images. In this study, we highlight this problem and address a novel task: universal deep image compression. This task aims to compress images belonging to arbitrary domains, such as natural images, line drawings, and comics. To address this problem, we propose a content-adaptive optimization framework; this framework uses a pre-trained compression model and adapts the model to a target image during compression. Adapters are inserted into the decoder of the model. For each input image, our framework optimizes the latent representation extracted by the encoder and the adapter parameters in terms of rate-distortion. The adapter parameters are additionally transmitted per image. For the experiments, a benchmark dataset containing uncompressed images of four domains (natural images, line drawings, comics, and vector arts) is constructed and the proposed universal deep compression is evaluated. Finally, the proposed model is compared with non-adaptive and existing adaptive compression models. The comparison reveals that the proposed model outperforms these. The code and dataset are publicly available at https://github.com/kktsubota/universal-dic.

In this work, we develop a theoretical and experimental framework to study the effect of margin penalties on angular softmax losses, which have led to state-of-the-art performance in face recognition. We also introduce a new multiplicative margin which performs comparably to previously proposed additive margins when the model is trained to convergence. A regime of the margin parameters can lead to degenerate minima, but these can be reliably avoided through the use of two regularization techniques that we propose. Our theory predicts the minimal angular distance between sample embeddings and the correct and wrong class prototype vectors learned during training, and it suggests a new method to identify optimal margin parameters without expensive tuning. Finally, we conduct a thorough ablation study of the margin parameters in our proposed framework, and we characterize the sensitivity of generalization to each parameter both theoretically and through experiments on standard face recognition benchmarks.

While remarkable progress has been made in imbalanced supervised learning, less attention has been given to the setting of imbalanced semi-supervised learning (SSL) where not only is few labeled data provided, but the underlying data distribution can be severely imbalanced. Recent work requires both complicated sampling strategies of pseudo-labeled unlabeled data and distribution alignment of the pseudo-label distribution to accommodate this imbalance. We present a novel approach that relies only on a form of a distribution alignment but no sampling strategy where rather than aligning the pseudo-labels during inference, we move the distribution alignment component into the respective cross entropy loss computations for both the supervised and unsupervised losses. This alignment compensates for both imbalance in the data and the eventual distributional shift present during evaluation. Altogether, this provides a unified strategy that offers both significantly reduced training requirements and improved performance across both low and richly labeled regimes and over varying degrees of imbalance. In experiments, we validate the efficacy of our method on SSL variants of CIFAR10-LT, CIFAR100-LT, and ImageNet-127. On ImageNet-127, our method shows 1.6% accuracy improvement over CReST with an 80% training time reduction and is competitive with other SOTA methods.

Mixup is a popular data augmentation technique based on creating new samples by linear interpolation between two given data samples, to improve both the generalization and robustness of the trained model. Knowledge distillation (KD), on the other hand, is widely used for model compression and transfer learning, which involves using a larger network's implicit knowledge to guide the learning of a smaller network. At first glance, these two techniques seem very different, however, we found that "smoothness" is the connecting link between the two and is also a crucial attribute in understanding KD's interplay with mixup. Although many mixup variants and distillation methods have been proposed, much remains to be understood regarding the role of a mixup in knowledge distillation. In this paper, we present a detailed empirical study on various important dimensions of compatibility between mixup and knowledge distillation. We also scrutinize the behavior of the networks trained with a mixup in the light of knowledge distillation through extensive analysis, visualizations, and comprehensive experiments on image classification. Finally, based on our findings, we suggest improved strategies to guide the student network to enhance its effectiveness. Additionally, the findings of this study provide insightful suggestions to researchers and practitioners that commonly use techniques from KD. Our code is available at https://github.com/hchoi71/MIX-KD.

Despite significant progress over the past few years, ambiguity is still a key challenge in Facial Expression Recognition (FER). It can lead to noisy and inconsistent annotation, which hinders the performance of deep learning models in real-world scenarios. In this paper, we propose a new uncertainty-aware label distribution learning method to improve the robustness of deep models against uncertainty and ambiguity. We leverage neighborhood information in the valence-arousal space to adaptively construct emotiona distributions for training samples. We also consider the uncertainty of provided labels when incorporating them into the label distributions. Our method can be easily integrated into a deep network to obtain more training supervision and improve recognition accuracy. Intensive experiments on several datasets under various noisy and ambiguous settings show that our method achieves competitive results and outperforms recent state-of-the-art approaches. Our code and models are available at https://github.com/minhnhatvt/label-distribution-learning-fer-tf.

Programmable scan LiDAR is able to measure arbitrary areas and is expected to be used in various applications. In this paper, we study a LiDAR sampling strategy for deep depth completion of a programmable scan LiDAR with an RGB camera. General data sampling strategies include adaptive approaches such as active learning, in which candidate data are assessed through a task model for data selection and then the selected data pool is updated sequentially. Although it is an effective approach, the adaptive approach requires many iterations involving the inference process to assess the candidate data, which is not suitable for LiDAR systems. Therefore, we propose a novel interactive LiDAR sampling method without each inference process. Our key insights are that we assess sampling candidates by depth estimation uncertainty and virtually update the uncertainty by an approximation of the candidate assessment. This enables us to add interactivity to the model state without requiring each inference process. We demonstrate the effectiveness of our method on the KITTI dataset and the generalization performance on the NYU-Depth-v2 dataset in comparison with a conventional adaptive LiDAR sampling method, and we find superior results in the depth completion task. We also show ablation studies to analyze our approach.

Unpaired image-to-image translation has broad applications in art, design, and scientific simulations. One early breakthrough was CycleGAN that emphasizes one-to-one mappings between two unpaired image domains via generative-adversarial networks (GAN) coupled with the cycle-consistency constraint, while more recent works promote one-to-many mapping to boost diversity of the translated images. Motivated by scientific simulation and one-to-one needs, this work revisits the classic CycleGAN framework and boosts its performance to outperform more contemporary models without relaxing the cycle-consistency constraint. To achieve this, we equip the generator with a Vision Transformer (ViT) and employ necessary training and regularization techniques. Compared to previous best-performing models, our model performs better and retains a strong correlation between the original and translated image. An accompanying ablation study shows that both the gradient penalty and self-supervised pre-training are crucial to the improvement. To promote reproducibility and open science, the source code, hyperparameter configurations, and pre-trained model are available at https: //github.com/LS4GAN/uvcgan.

Recognizing soft-biometric pedestrian attributes is essential in video surveillance and fashion retrieval. Recent works show promising results on single datasets. Nevertheless, the generalization ability of these methods under different attribute distributions, viewpoints, varying illumination, and low resolutions remains rarely understood due to strong biases and varying attributes in current datasets. To close this gap and support a systematic investigation, we present UPAR, the Unified Person Attribute Recognition Dataset. It is based on four well-known person attribute recognition datasets: PA100K, PETA, RAPv2, and Market1501. We unify those datasets by providing 3,3M additional annotations to harmonize 40 important binary attributes over 12 attribute categories across the datasets. We thus enable research on generalizable pedestrian attribute recognition as well as attribute-based person retrieval for the first time. Due to the vast variance of the image distribution, pedestrian pose, scale, and occlusion, existing approaches are greatly challenged both in terms of accuracy and efficiency. Furthermore, we develop a strong baseline for PAR and attribute-based person retrieval based on a thorough analysis of regularization methods. Our models achieve state-of-the-art performance in cross-domain and specialization settings on PA100k, PETA, RAPv2, Market1501-Attributes, and UPAR. We believe UPAR and our strong baseline will contribute to the artificial intelligence community and promote research on large-scale, generalizable attribute recognition systems.