CVPR 2021 论文列表

We propose real-time, six degrees of freedom (6DoF), 3D face pose estimation without face detection or landmark localization. We observe that estimating the 6DoF rigid transformation of a face is a simpler problem than facial landmark detection, often used for 3D face alignment. In addition, 6DoF offers more information than face bounding box labels. We leverage these observations to make multiple contributions: (a) We describe an easily trained, efficient, Faster R-CNN-based model which regresses 6DoF pose for all faces in the photo, without preliminary face detection. (b) We explain how pose is converted and kept consistent between the input photo and arbitrary crops created while training and evaluating our model. (c) Finally, we show how face poses can replace detection bounding box training labels. Tests on AFLW2000-3D and BIWI show that our method runs at real-time and outperforms state of the art (SotA) face pose estimators. Remarkably, our method also surpasses SotA models of comparable complexity on the WIDER FACE detection benchmark, despite not been optimized on bounding box labels.

It is nontrivial to store rapidly growing big data nowadays, which demands high-performance lossless compression techniques. Likelihood-based generative models have witnessed their success on lossless compression, where flow based models are desirable in allowing exact data likelihood optimisation with bijective mappings. However, common continuous flows are in contradiction with the discreteness of coding schemes, which requires either 1) imposing strict constraints on flow models that degrades the performance or 2) coding numerous bijective mapping errors which reduces the efficiency. In this paper, we investigate volume preserving flows for lossless compression and show that a bijective mapping without error is possible. We propose Numerical Invertible Volume Preserving Flow (iVPF) which is derived from the general volume preserving flows. By introducing novel computation algorithms on flow models, an exact bijective mapping is achieved without any numerical error. We also propose a lossless compression algorithm based on iVPF. Experiments on various datasets show that the algorithm based on iVPF achieves state-of-the-art compression ratio over lightweight compression algorithms.

We introduce a new dataset for the emotional artificial intelligence research: identity-free video dataset for micro-gesture understanding and emotion analysis (iMiGUE). Different from existing public datasets, iMiGUE focuses on nonverbal body gestures without using any identity information, while the predominant researches of emotion analysis concern sensitive biometric data, like face and speech. Most importantly, iMiGUE focuses on micro-gestures, i,e., unintentional behaviors driven by inner feelings, which are different from ordinary scope of gestures from other gesture datasets which are mostly intentionally performed for illustrative purposes. Furthermore, iMiGUE is designed to evaluate the ability of models to analyze the emotional states by integrating information of recognized micro-gesture, rather than just recognizing prototypes in the sequences separately (or isolatedly). This is because the real need for emotion AI is to understand the emotional states behind gestures in a holistic way. Moreover, to counter for the challenge of imbalanced samples distribution of this dataset, an unsupervised learning method is proposed to capture latent representations from the micro-gesture sequences themselves. We systematically investigate representative methods on this dataset, and comprehensive experimental results reveal several interesting insights from the iMiGUE, e,g., micro-gesture-based analysis can promote emotion understanding. We confirm that the new iMiGUE dataset could advance studies of micro-gesture and emotion AI.

We present the first deep implicit 3D morphable model (i3DMM) of full heads. Unlike earlier morphable face models it not only captures identity-specific geometry, texture, and expressions of the frontal face, but also models the entire head, including hair. We collect a new dataset consisting of 64 people with different expressions and hairstyles to train i3DMM. Our approach has the following favorable properties: (i) It is the first full head morphable model that includes hair. (ii) In contrast to mesh-based models it can be trained on merely rigidly aligned scans, without requiring difficult non-rigid registration. (iii) We design a novel architecture to decouple the shape model into an implicit reference shape and a deformation of this reference shape. With that, dense correspondences between shapes can be learned implicitly. (iv) This architecture allows us to semantically disentangle the geometry and color components, as color is learned in the reference space. Geometry is further disentangled as identity, expressions, and hairstyle, while color is disentangled as identity and hairstyle components. We show the merits of i3DMM using ablation studies, comparisons to state-of-the-art models, and applications such as semantic head editing and texture transfer. We will make our model publicly available.

Accurate segmentation of tubular, network-like structures, such as vessels, neurons, or roads, is relevant to many fields of research. For such structures, the topology is their most important characteristic; particularly preserving connectedness: in the case of vascular networks, missing a connected vessel entirely alters the blood-flow dynamics. We introduce a novel similarity measure termed centerlineDice (short clDice), which is calculated on the intersection of the segmentation masks and their (morphological) skeleta. We theoretically prove that clDice guarantees topology preservation up to homotopy equivalence for binary 2D and 3D segmentation. Extending this, we propose a computationally efficient, differentiable loss function (soft-clDice) for training arbitrary neural segmentation networks. We benchmark the soft-clDice loss on five public datasets, including vessels, roads and neurons (2D and 3D). Training on soft-clDice leads to segmentation with more accurate connectivity information, higher graph similarity, and better volumetric scores.

We present Zillow Indoor Dataset (ZInD): A large indoor dataset with 71,474 panoramas from 1,524 real unfurnished homes. ZInD provides annotations of 3D room layouts, 2D and 3D floor plans, panorama location in the floor plan, and locations of windows and doors. The ground truth construction took over 1,500 hours of annotation work. To the best of our knowledge, ZInD is the largest real dataset with layout annotations. A unique property is the room layout data, which follows a real world distribution (cuboid, more general Manhattan, and non-Manhattan layouts) as opposed to the mostly cuboid or Manhattan layouts in current publicly available datasets. Also, the scale and annotations provided are valuable for effective research related to room layout and floor plan analysis. To demonstrate ZInD's benefits, we benchmark on room layout estimation from single panoramas and multi-view registration.

Adverse weather conditions, including snow, rain, and fog, pose a major challenge for both human and computer vision. Handling these environmental conditions is essential for safe decision making, especially in autonomous vehicles, robotics, and drones. Most of today's supervised imaging and vision approaches, however, rely on training data collected in the real world that is biased towards good weather conditions, with dense fog, snow, and heavy rain as outliers in these datasets. Without training data, let alone paired data, existing autonomous vehicles often limit themselves to good conditions and stop when dense fog or snow is detected. In this work, we tackle the lack of supervised training data by combining synthetic and indirect supervision. We present ZeroScatter, a domain transfer method for converting RGB-only captures taken in adverse weather into clear daytime scenes. ZeroScatter exploits model-based, temporal, multi-view, multi-modal, and adversarial cues in a joint fashion, allowing us to train on unpaired, biased data. We assess the proposed method on in-the-wild captures, and the proposed method outperforms existing monocular descattering approaches by 2.8 dB PSNR on controlled fog chamber measurements.

Real-world image degradation due to light scattering can be described based on the Koschmieder's model. Training deep models to restore such degraded images is challenging as real-world paired data is scarcely available and synthetic paired data may suffer from domain-shift issues. In this paper, a zero-shot single real-world image restoration model is proposed leveraging a theoretically deduced property of degradation through the Koschmieder's model. Our zero-shot network estimates the parameters of the Koschmieder's model, which describes the degradation in the input image, to perform image restoration. We show that a suitable degradation of the input image amounts to a controlled perturbation of the Koschmieder's model that describes the image's formation. The optimization of the zero-shot network is achieved by seeking to maintain the relation between its estimates of Koschmieder's model parameters before and after the controlled perturbation, along with the use of a few no-reference losses. Image dehazing and underwater image restoration are carried out using the proposed zero-shot framework, which in general outperforms the state-of-the-art quantitatively and subjectively on multiple standard real-world image datasets. Additionally, the application of our zero-shot framework for low-light image enhancement is also demonstrated.

Deep learning has significantly improved the precision of instance segmentation with abundant labeled data. However, in many areas like medical and manufacturing, collecting sufficient data is extremely hard and labeling this data requires high professional skills. We follow this motivation and propose a new task set named zero-shot instance segmentation (ZSI). In the training phase of ZSI, the model is trained with seen data, while in the testing phase, it is used to segment all seen and unseen instances. We first formulate the ZSI task and propose a method to tackle the challenge, which consists of Zero-shot Detector, Semantic Mask Head, Background Aware RPN and Synchronized Background Strategy. We present a new benchmark for zero-shot instance segmentation based on the MS-COCO dataset. The extensive empirical results in this benchmark show that our method not only surpasses the state-of-the-art results in zero-shot object detection task but also achieves promising performance on ZSI. Our approach will serve as a solid baseline and facilitate future research in zero-shot instance segmentation. Code available at ZSI.

Model quantization is a promising approach to compress deep neural networks and accelerate inference, making it possible to be deployed on mobile and edge devices. To retain the high performance of full-precision models, most existing quantization methods focus on fine-tuning quantized model by assuming training datasets are accessible. However, this assumption sometimes is not satisfied in real situations due to data privacy and security issues, thereby making these quantization methods not applicable. To achieve zero-short model quantization without accessing training data, a tiny number of quantization methods adopt either post-training quantization or batch normalization statistics-guided data generation for fine-tuning. However, both of them inevitably suffer from low performance, since the former is a little too empirical and lacks training support for ultra-low precision quantization, while the latter could not fully restore the peculiarities of original data and is often low efficient for diverse data generation. To address the above issues, we propose a zero-shot adversarial quantization (ZAQ) framework, facilitating effective discrepancy estimation and knowledge transfer from a full-precision model to its quantized model. This is achieved by a novel two-level discrepancy modeling to drive a generator to synthesize informative and diverse data examples to optimize the quantized model in an adversarial learning fashion. We conduct extensive experiments on three fundamental vision tasks, demonstrating the superiority of ZAQ over the strong zero-shot baselines and validating the effectiveness of its main components.

Whether what you see in Figure 1 is a "flamingo" or a "bird", is the question we ask in this paper. While fine-grained visual classification (FGVC) strives to arrive at the former, for the majority of us non-experts just "bird" would probably suffice. The real question is therefore -- how can we tailor for different fine-grained definitions under divergent levels of expertise. For that, we re-envisage the traditional setting of FGVC, from single-label classification, to that of top-down traversal of a pre-defined coarse-to-fine label hierarchy -- so that our answer becomes "bird"="Phoenicopteriformes"="Phoenicopteridae"="flamingo". To approach this new problem, we first conduct a comprehensive human study where we confirm that most participants prefer multi-granularity labels, regardless whether they consider themselves experts. We then discover the key intuition that: coarse-level label prediction exacerbates fine-grained feature learning, yet fine-level feature betters the learning of coarse-level classifier. This discovery enables us to design a very simple albeit surprisingly effective solution to our new problem, where we (i) leverage level-specific classification heads to disentangle coarse-level features with fine-grained ones, and (ii) allow finer-grained features to participate in coarser-grained label predictions, which in turn helps with better disentanglement. Experiments show that our method achieves superior performance in the new FGVC setting, and performs better than state-of-the-art on traditional single-label FGVC problem as well. Thanks to its simplicity, our method can be easily implemented on top of any existing FGVC frameworks and is parameter-free. Codes are available at: https://github.com/PRIS-CV/Fine-Grained-or-Not

With the large multimedia content online, deep hashing has become a popular method for efficient image retrieval and storage. However, by inheriting the algorithmic backend from softmax classification, these techniques are vulnerable to the well-known adversarial examples as well. The massive collection of online images into the database also opens up new attack vectors. Attackers can embed adversarial images into the database and target specific categories to be retrieved by user queries. In this paper, we start from an adversarial standpoint to explore and enhance the capacity of targeted black-box transferability attack for deep hashing. We motivate this work by a series of empirical studies to see the unique challenges in image retrieval. We study the relations between adversarial subspace and black-box transferability via utilizing random noise as a proxy. Then we develop a new attack that is simultaneously adversarial and robust to noise to enhance transferability. Our experimental results demonstrate about 1.2-3x improvements of black-box transferability compared with the state-of-the-art mechanisms.

This paper revisits feature pyramids networks (FPN) for one-stage detectors and points out that the success of FPN is due to its divide-and-conquer solution to the optimization problem in object detection rather than multi-scale feature fusion. From the perspective of optimization, we introduce an alternative way to address the problem instead of adopting the complex feature pyramids -- utilizing only one-level feature for detection. Based on the simple and efficient solution, we present You Only Look One-level Feature (YOLOF). In our method, two key components, Dilated Encoder and Uniform Matching, are proposed and bring considerable improvements. Extensive experiments on the COCO benchmark prove the effectiveness of the proposed model. Our YOLOF achieves comparable results with its feature pyramids counterpart RetinaNet while being 2.5 times faster. Without transformer layers, YOLOF can match the performance of DETR in a single-level feature manner with 7 times less training epochs.

Automated diagnosis using deep neural networks in chest radiography can help radiologists detect life-threatening diseases. However, existing methods only provide predictions without accurate explanations, undermining the trustworthiness of the diagnostic methods. Here, we present XProtoNet, a globally and locally interpretable diagnosis framework for chest radiography. XProtoNet learns representative patterns of each disease from X-ray images, which are prototypes, and makes a diagnosis on a given X-ray image based on the patterns. It predicts the area where a sign of the disease is likely to appear and compares the features in the predicted area with the prototypes. It can provide a global explanation, the prototype, and a local explanation, how the prototype contributes to the prediction of a single image. Despite the constraint for interpretability, XProtoNet achieves state-of-the-art classification performance on the public NIH chest X-ray dataset.

6D pose estimation in space poses unique challenges that are not commonly encountered in the terrestrial setting. One of the most striking differences is the lack of atmospheric scattering, allowing objects to be visible from a great distance while complicating illumination conditions. Currently available benchmark datasets do not place a sufficient emphasis on this aspect and mostly depict the target in close proximity. Prior work tackling pose estimation under large scale variations relies on a two-stage approach to first estimate scale, followed by pose estimation on a resized image patch. We instead propose a single-stage hierarchical end-to-end trainable network that is more robust to scale variations. We demonstrate that it outperforms existing approaches not only on images synthesized to resemble images taken in space but also on standard benchmarks.

Modern deep learning techniques that regress the relative camera pose between two images have difficulty dealing with challenging scenarios, such as large camera motions resulting in occlusions and significant changes in perspective that leave little overlap between images. These models continue to struggle even with the benefit of large supervised training datasets. To address the limitations of these models, we take inspiration from techniques that show regressing keypoint locations in 2D and 3D can be improved by estimating a discrete distribution over keypoint locations. Analogously, in this paper we explore improving camera pose regression by instead predicting a discrete distribution over camera poses. To realize this idea, we introduce DirectionNet, which estimates discrete distributions over the 5D relative pose space using a novel parameterization to make the estimation problem tractable. Specifically, DirectionNet factorizes relative camera pose, specified by a 3D rotation and a translation direction, into a set of 3D direction vectors. Since 3D directions can be identified with points on the sphere, DirectionNet estimates discrete distributions on the sphere as its output. We evaluate our model on challenging synthetic and real pose estimation datasets constructed from Matterport3D and InteriorNet. Promising results show a near 50% reduction in error over direct regression methods.

We address the problem of estimating the 3D pose of a network of cameras for large-environment wide-baseline scenarios, e.g., cameras for construction sites, sports stadiums, and public spaces. This task is challenging since detecting and matching the same 3D keypoint observed from two very different camera views is difficult, making standard structure-from-motion (SfM) pipelines inapplicable. In such circumstances, treating people in the scene as "keypoints" and associating them across different camera views can be an alternative method for obtaining correspondences. Based on this intuition, we propose a method that uses ideas from person re-identification (re-ID) for wide-baseline camera calibration. Our method first employs a re-ID method to associate human bounding boxes across cameras, then converts bounding box correspondences to point correspondences, and finally solves for camera pose using multi-view geometry and bundle adjustment. Since our method does not require specialized calibration targets except for visible people, it applies to situations where frequent calibration updates are required. We perform extensive experiments on datasets captured from scenes of different sizes, camera settings (indoor and outdoor), and human activities (walking, playing basketball, construction). Experiment results show that our method achieves similar performance to standard SfM methods relying on manually labeled point correspondences.

Capturing interpretable variations has long been one of the goals in disentanglement learning. However, unlike the independence assumption, interpretability has rarely been exploited to encourage disentanglement in the unsupervised setting. In this paper, we examine the interpretability of disentangled representations by investigating two questions: where to be interpreted and what to be interpreted? A latent code is easily to be interpreted if it would consistently impact a certain subarea of the resulting generated image. We thus propose to learn a spatial mask to localize the effect of each individual latent dimension. On the other hand, interpretability usually comes from latent dimensions that capture simple and basic variations in data. We thus impose a perturbation on a certain dimension of the latent code, and expect to identify the perturbation along this dimension from the generated images so that the encoding of simple variations can be enforced. Additionally, we develop an unsupervised model selection method, which accumulates perceptual distance scores along axes in the latent space. On various datasets, our models can learn high-quality disentangled representations without supervision, showing the proposed modeling of interpretability is an effective proxy for achieving unsupervised disentanglement.

Human pose estimation is a fundamental yet challenging task in computer vision, which aims at localizing human anatomical keypoints. However, unlike human vision that is robust to various data corruptions such as blur and pixelation, current pose estimators are easily confused by these corruptions. This work comprehensively studies and addresses this problem by building rigorous robust benchmarks, termed COCO-C, MPII-C, and OCHuman-C, to evaluate the weaknesses of current advanced pose estimators, and a new algorithm termed AdvMix is proposed to improve their robustness in different corruptions. Our work has several unique benefits. (1) AdvMix is model-agnostic and capable in a wide-spectrum of pose estimation models. (2) AdvMix consists of adversarial augmentation and knowledge distillation. Adversarial augmentation contains two neural network modules that are trained jointly and competitively in an adversarial manner, where a generator network mixes different corrupted images to confuse a pose estimator, improving the robustness of the pose estimator by learning from harder samples. To compensate for the noise patterns by adversarial augmentation, knowledge distillation is applied to transfer clean pose structure knowledge to the target pose estimator. (3) Extensive experiments show that AdvMix significantly increases the robustness of pose estimations across a wide range of corruptions, while maintaining accuracy on clean data in various challenging benchmark datasets.