CVPR 2020 Papers

Although automatic gaze estimation is very important to a large variety of application areas, it is difficult to train accurate and robust gaze models, in great part due to the difficulty in collecting large and diverse data (annotating 3D gaze is expensive and existing datasets use different setups). To address this issue, our main contribution in this paper is to propose an effective approach to learn a low dimensional gaze representation without gaze annotations, which to the best of our best knowledge, is the first work to do so. The main idea is to rely on a gaze redirection network and use the gaze representation difference of the input and target images (of the redirection network) as the redirection variable. A redirection loss in image domain allows the joint training of both the redirection network and the gaze representation network. In addition, we propose a warping field regularization which not only provides an explicit physical meaning to the gaze representations but also avoids redirection distortions. Promising results on few-shot gaze estimation (competitive results can be achieved with as few as <= 100 calibration samples), cross-dataset gaze estimation, gaze network pretraining, and another task (head pose estimation) demonstrate the validity of our framework.

Visual navigation is a task of training an embodied agent by intelligently navigating to a target object (e.g., television) using only visual observations. A key challenge for current deep reinforcement learning models lies in the requirements for a large amount of training data. It is exceedingly expensive to construct sufficient 3D synthetic environments annotated with the target object information. In this paper, we focus on visual navigation in the low-resource setting, where we have only a few training environments annotated with object information. We propose a novel unsupervised reinforcement learning approach to learn transferable meta-skills (e.g., bypass obstacles, go straight) from unannotated environments without any supervisory signals. The agent can then fast adapt to visual navigation through learning a high-level master policy to combine these meta-skills, when the visual-navigation-specified reward is provided. Experimental results show that our method significantly outperforms the baseline by 53.34% relatively on SPL, and further qualitative analysis demonstrates that our method learns transferable motor primitives for visual navigation.

Person re-identification (re-ID) is an important topic in computer vision. This paper studies the unsupervised setting of re-ID, which does not require any labeled information and thus is freely deployed to new scenarios. There are very few studies under this setting, and one of the best approach till now used iterative clustering and classification, so that unlabeled images are clustered into pseudo classes for a classifier to get trained, and the updated features are used for clustering and so on. This approach suffers two problems, namely, the difficulty of determining the number of clusters, and the hard quantization loss in clustering. In this paper, we follow the iterative training mechanism but discard clustering, since it incurs loss from hard quantization, yet its only product, image-level similarity, can be easily replaced by pairwise computation and a softened classification task. With these improvements, our approach becomes more elegant and is more robust to hyper-parameter changes. Experiments on two image-based and video-based datasets demonstrate state-of-the-art performance under the unsupervised re-ID setting.

Many applications, such as autonomous driving, heavily rely on multi-modal data where spatial alignment between the modalities is required. Most multi-modal registration methods struggle computing the spatial correspondence between the images using prevalent cross-modality similarity measures. In this work, we bypass the difficulties of developing cross-modality similarity measures, by training an image-to-image translation network on the two input modalities. This learned translation allows training the registration network using simple and reliable mono-modality metrics. We perform multi-modal registration using two networks - a spatial transformation network and a translation network. We show that by encouraging our translation network to be geometry preserving, we manage to train an accurate spatial transformation network. Compared to state-of-the-art multi-modal methods our presented method is unsupervised, requiring no pairs of aligned modalities for training, and can be adapted to any pair of modalities. We evaluate our method quantitatively and qualitatively on commercial datasets, showing that it performs well on several modalities and achieves accurate alignment.

This work investigates the task of unsupervised model personalization, adapted to continually evolving, unlabeled local user images. We consider the practical scenario where a high capacity server interacts with a myriad of resource-limited edge devices, imposing strong requirements on scalability and local data privacy. We aim to address this challenge within the continual learning paradigm and provide a novel Dual User-Adaptation framework (DUA) to explore the problem. This framework flexibly disentangles user-adaptation into model personalization on the server and local data regularization on the user device, with desirable properties regarding scalability and privacy constraints. First, on the server, we introduce incremental learning of task-specific expert models, subsequently aggregated using a concealed unsupervised user prior. Aggregation avoids retraining, whereas the user prior conceals sensitive raw user data, and grants unsupervised adaptation. Second, local user-adaptation incorporates a domain adaptation point of view, adapting regularizing batch normalization parameters to the user data. We explore various empirical user configurations with different priors in categories and a tenfold of transforms for MIT Indoor Scene recognition, and classify numbers in a combined MNIST and SVHN setup. Extensive experiments yield promising results for data-driven local adaptation and elicit user priors for server adaptation to depend on the model rather than user data. Hence, although user-adaptation remains a challenging open problem, the DUA framework formalizes a principled foundation for personalizing both on server and user device, while maintaining privacy and scalability.

Analyzing human posture and precisely comparing it across different subjects is essential for accurate understanding of behavior and numerous vision applications such as medical diagnostics, sports, or surveillance. Motion magnification techniques help to see even small deviations in posture that are invisible to the naked eye. However, they fail when comparing subtle posture differences across individuals with diverse appearance. Keypoint-based posture estimation and classification techniques can handle large variations in appearance, but are invariant to subtle deviations in posture. We present an approach to unsupervised magnification of posture differences across individuals despite large deviations in appearance. We do not require keypoint annotation and visualize deviations on a sub-bodypart level. To transfer appearance across subjects onto a magnified posture, we propose a novel loss for disentangling appearance and posture in an autoencoder. Posture magnification yields exaggerated images that are different from the training set. Therefore, we incorporate magnification already into the training of the disentangled autoencoder and learn on real data and synthesized magnifications without supervision. Experiments confirm that our approach improves upon the state-of-the-art in magnification and on the application of discovering posture deviations due to impairment.

Learning structures of 3D shapes is a fundamental problem in the field of computer graphics and geometry processing. We present a simple yet interpretable unsupervised method for learning a new structural representation in the form of 3D structure points. The 3D structure points produced by our method encode the shape structure intrinsically and exhibit semantic consistency across all the shape instances with similar structures. This is a challenging goal that has not fully been achieved by other methods. Specifically, our method takes a 3D point cloud as input and encodes it as a set of local features. The local features are then passed through a novel point integration module to produce a set of 3D structure points. The chamfer distance is used as reconstruction loss to ensure the structure points lie close to the input point cloud. Extensive experiments have shown that our method outperforms the state-of-the-art on the semantic shape correspondence task and achieves comparable performance with the state-of-the-art on the segmentation label transfer task. Moreover, the PCA based shape embedding built upon consistent structure points demonstrates good performance in preserving the shape structures. Code is available at https://github.com/NolenChen/3DStructurePoints

Intrinsic image decomposition, which is an essential task in computer vision, aims to infer the reflectance and shading of the scene. It is challenging since it needs to separate one image into two components. To tackle this, conventional methods introduce various priors to constrain the solution, yet with limited performance. Meanwhile, the problem is typically solved by supervised learning methods, which is actually not an ideal solution since obtaining ground truth reflectance and shading for massive general natural scenes is challenging and even impossible. In this paper, we propose a novel unsupervised intrinsic image decomposition framework, which relies on neither labeled training data nor hand-crafted priors. Instead, it directly learns the latent feature of reflectance and shading from unsupervised and uncorrelated data. To enable this, we explore the independence between reflectance and shading, the domain invariant content constraint and the physical constraint. Extensive experiments on both synthetic and real image datasets demonstrate consistently superior performance of the proposed method.

Because of the rich dynamical structure of videos andtheir ubiquity in everyday life, it is a natural idea that video data could serve as a powerful unsupervised learning signal for visual representations. However, instantiating this idea, especially at large scale, has remained a significant artificial intelligence challenge. Here we present the Video Instance Embedding (VIE) framework, which trains deep nonlinear embeddings on video sequence inputs. By learning embedding dimensions that identify and group similar videos together, while pushing inherently different videos apart in the embedding space, VIE captures the strong statistical structure inherent in videos, without the need for external annotation labels. We find that, when trained on a large-scale video dataset, VIE yields powerful representations both for action recognition and single-frame object categorization, showing substantially improving on the state of the art wherever direct comparisons are possible. We show that a two-pathway model with both static and dynamic processingpathways is optimal, provide analyses indicating how the model works, and perform ablation studies showing the importance of key architecture and loss function choices. Our results suggest that deep neural embeddings are a promising approach to unsupervised video learning fora wide variety of task domains.

Unsupervised domain adaptation (UDA) for nuclei instance segmentation is important for digital pathology, as it alleviates the burden of labor-intensive annotation and domain shift across datasets. In this work, we propose a Cycle Consistency Panoptic Domain Adaptive Mask R-CNN (CyC-PDAM) architecture for unsupervised nuclei segmentation in histopathology images, by learning from fluorescence microscopy images. More specifically, we first propose a nuclei inpainting mechanism to remove the auxiliary generated objects in the synthesized images. Secondly, a semantic branch with a domain discriminator is designed to achieve panoptic-level domain adaptation. Thirdly, in order to avoid the influence of the source-biased features, we propose a task re-weighting mechanism to dynamically add trade-off weights for the task-specific loss functions. Experimental results on three datasets indicate that our proposed method outperforms state-of-the-art UDA methods significantly, and demonstrates a similar performance as fully supervised methods.

Domain adaptation attempts to boost the performance on a target domain by borrowing knowledge from a well established source domain. To handle the distribution gap between two domains, the prominent approaches endeavor to extract domain-invariant features. It is known that after a perfect domain alignment the domain-invariant representations of two domains should share the same characteristics from perspective of the overview and also any local piece. Inspired by this, we propose a novel method called Hierarchical Gradient Synchronization to model the synchronization relationship among the local distribution pieces and global distribution, aiming for more precise domain-invariant features. Specifically, the hierarchical domain alignments including class-wise alignment, group-wise alignment and global alignment are first constructed. Then, these three types of alignment are constrained to be consistent to ensure better structure preservation. As a result, the obtained features are domain invariant and intrinsically structure preserved. As evaluated on extensive domain adaptation tasks, our proposed method achieves state-of-the-art classification performance on both vanilla unsupervised domain adaptation and partial domain adaptation.

Deep learning models have achieved great success in supervised shape descriptor learning for 3D shape retrieval, classification, and correspondence. However, the unsupervised shape descriptor calculated via deep learning is less studied than that of supervised ones due to the design challenges of unsupervised neural network architecture. This paper proposes a novel probabilistic framework for the learning of unsupervised deep shape descriptors with point distribution learning. In our approach, we firstly associate each point with a Gaussian, and the point clouds are modeled as the distribution of the points. We then use deep neural networks (DNNs) to model a maximum likelihood estimation process that is traditionally solved with an iterative Expectation-Maximization (EM) process. Our key novelty is that "training" these DNNs with unsupervised self-correspondence L2 distance loss will elegantly reveal the statically significant deep shape descriptor representation for the distribution of the point clouds. We have conducted experiments over various 3D datasets. Qualitative and quantitative comparisons demonstrate that our proposed method achieves superior classification performance over existing unsupervised 3D shape descriptors. In addition, we verified the following attractive properties of our shape descriptor through experiments: multi-scale shape representation, robustness to shape rotation, and robustness to noise.

The key for fusion based hyperspectral image (HSI) super-resolution (SR) is to infer the posteriori of a latent HSI using appropriate image prior and likelihood that depends on degeneration. However, in practice the priors of high-dimensional HSIs can be extremely complicated and the degeneration is often unknown. Consequently most existing approaches that assume a shallow hand-crafted image prior and a pre-defined degeneration, fail to well generalize in real applications. To tackle this problem, we present an unsupervised adaptation learning (UAL) framework. Instead of directly modelling the complicated image prior, we propose to first implicitly learn a general image prior using deep networks and then adapt it to a specific HSI. Following this idea, we develop a two-stage SR network that leverages two consecutive modules: a fusion module and an adaptation module, to recover the latent HSI in a coarse-to-fine scheme. The fusion module is pretrained in a supervised manner on synthetic data to capture a spatial-spectral prior that is general across most HSIs. To adapt the learned general prior to the specific HSI under unknown degeneration, we introduce a simple degeneration network to assist learning both the adaptation module and the degeneration in an unsupervised way. In this way, the resultant image-specific prior and the estimated degeneration can benefit the inference of a more accurate posteriori, thereby increasing generalization capacity. To verify the efficacy of UAL, we extensively evaluate it on four benchmark datasets and report strong results that surpass existing approaches.

Portrait drawing is a common form of art with high abstraction and expressiveness. Due to its unique characteristics, existing methods achieve decent results only with paired training data, which is costly and time-consuming to obtain.In this paper, we address the problem of automatic transfer from face photos to portrait drawings with unpaired training data. We observe that due to the significant imbalance of information richness between photos and drawings, existing unpaired transfer methods such as CycleGAN tends to embed invisible reconstruction information indiscriminately in the whole drawings, leading to important facial features partially missing in drawings. To address this problem, we propose a novel asymmetric cycle mapping that enforces the reconstruction information to be visible (by a truncation loss) and only embedded in selective facial regions (by a relaxed forward cycle-consistency loss). Along with localized discriminators for the eyes, nose and lips, our method well preserves all important facial features in the generated portrait drawings. By introducing a style classifier and taking the style vector into account, our method can learn to generate portrait drawings in multiple styles using a single network. Extensive experiments show that our model outperforms state-of-the-art methods.

Cross-modal matching has been a highlighted research topic in both vision and language areas. Learning appropriate mining strategy to sample and weight informative pairs is crucial for the cross-modal matching performance. However, most existing metric learning methods are developed for unimodal matching, which is unsuitable for cross-modal matching on multimodal data with heterogeneous features. To address this problem, we propose a simple and interpretable universal weighting framework for cross-modal matching, which provides a tool to analyze the interpretability of various loss functions. Furthermore, we introduce a new polynomial loss under the universal weighting framework, which defines a weight function for the positive and negative informative pairs respectively. Experimental results on two image-text matching benchmarks and two video-text matching benchmarks validate the efficacy of the proposed method.

There is a strong incentive to develop versatile learning techniques that can transfer the knowledge of class-separability from a labeled source domain to an unlabeled target domain in the presence of a domain-shift. Existing domain adaptation (DA) approaches are not equipped for practical DA scenarios as a result of their reliance on the knowledge of source-target label-set relationship (e.g. Closed-set, Open-set or Partial DA). Furthermore, almost all prior unsupervised DA works require coexistence of source and target samples even during deployment, making them unsuitable for real-time adaptation. Devoid of such impractical assumptions, we propose a novel two-stage learning process. 1) In the Procurement stage, we aim to equip the model for future source-free deployment, assuming no prior knowledge of the upcoming category-gap and domain-shift. To achieve this, we enhance the model's ability to reject out-of-source distribution samples by leveraging the available source data, in a novel generative classifier framework. 2) In the Deployment stage, the goal is to design a unified adaptation algorithm capable of operating across a wide range of category-gaps, with no access to the previously seen source samples. To this end, in contrast to the usage of complex adversarial training regimes, we define a simple yet effective source-free adaptation objective by utilizing a novel instance-level weighting mechanism, named as Source Similarity Metric (SSM). A thorough evaluation shows the practical usability of the proposed learning framework with superior DA performance even over state-of-the-art source-dependent approaches.

In this paper, we study physical adversarial attacks on object detectors in the wild. Previous works mostly craft instance-dependent perturbations only for rigid or planar objects. To this end, we propose to learn an adversarial pattern to effectively attack all instances belonging to the same object category, referred to as Universal Physical Camouflage Attack (UPC). Concretely, UPC crafts camouflage by jointly fooling the region proposal network, as well as misleading the classifier and the regressor to output errors. In order to make UPC effective for non-rigid or non-planar objects, we introduce a set of transformations for mimicking deformable properties. We additionally impose optimization constraint to make generated patterns look natural to human observers. To fairly evaluate the effectiveness of different physical-world attacks, we present the first standardized virtual database, AttackScenes, which simulates the real 3D world in a controllable and reproducible environment. Extensive experiments suggest the superiority of our proposed UPC compared with existing physical adversarial attackers not only in virtual environments (AttackScenes), but also in real-world physical environments.

Style variation has been a major challenge for person re-identification, which aims to match the same pedestrians across different cameras. Existing works attempted to address this problem with camera-invariant descriptor subspace learning. However, there will be more image artifacts when the difference between the images taken by different cameras is larger. To solve this problem, we propose a UnityStyle adaption method, which can smooth the style disparities within the same camera and across different cameras. Specifically, we firstly create UnityGAN to learn the style changes between cameras, producing shape-stable style-unity images for each camera, which is called UnityStyle images. Meanwhile, we use UnityStyle images to eliminate style differences between different images, which makes a better match between query and gallery. Then, we apply the proposed method to Re-ID models, expecting to obtain more style-robust depth features for querying. We conduct extensive experiments on widely used benchmark datasets to evaluate the performance of the proposed framework, the results of which confirm the superiority of the proposed model.

We introduce a powerful student-teacher framework for the challenging problem of unsupervised anomaly detection and pixel-precise anomaly segmentation in high-resolution images. Student networks are trained to regress the output of a descriptive teacher network that was pretrained on a large dataset of patches from natural images. This circumvents the need for prior data annotation. Anomalies are detected when the outputs of the student networks differ from that of the teacher network. This happens when they fail to generalize outside the manifold of anomaly-free training data. The intrinsic uncertainty in the student networks is used as an additional scoring function that indicates anomalies. We compare our method to a large number of existing deep learning based methods for unsupervised anomaly detection. Our experiments demonstrate improvements over state-of-the-art methods on a number of real-world datasets, including the recently introduced MVTec Anomaly Detection dataset that was specifically designed to benchmark anomaly segmentation algorithms.