ICCV 2021 论文列表

In this paper, we propose a generalizable mixed-precision quantization (GMPQ) method for efficient inference. Conventional methods require the consistency of datasets for bitwidth search and model deployment to guarantee the policy optimality, leading to heavy search cost on challenging largescale datasets in realistic applications. On the contrary, our GMPQ searches the mixed-quantization policy that can be generalized to largescale datasets with only a small amount of data, so that the search cost is significantly reduced without performance degradation. Specifically, we observe that locating network attribution correctly is general ability for accurate visual analysis across different data distribution. Therefore, despite of pursuing higher model accuracy and complexity, we preserve attribution rank consistency between the quantized models and their full-precision counterparts via efficient capacity-aware attribution imitation for generalizable mixed-precision quantization strategy search. Extensive experiments show that our method obtains competitive accuracy-complexity trade-off compared with the state-of-the-art mixed-precision networks in significantly reduced search cost. The code is available at https://github.com/ZiweiWangTHU/GMPQ.git.

Reconstructing delicate geometric details with consumer RGB-D sensors is challenging due to sensor depth and poses uncertainties. To tackle this problem, we propose a unique geometry-guided fusion framework: 1) First, we characterize fusion correspondences with the geodesic curves derived from the mass transport problem, also known as the Monge-Kantorovich problem. Compared with the depth map back-projection methods, the geodesic curves reveal the geometric structures of the local surface. 2) Moving the points along the geodesic curves is the core of our fusion approach, guided by local geometric properties, i.e., Gaussian curvature and mean curvature. Compared with the state-of-the-art methods, our novel geometry-guided displacement interpolation fully utilizes the meaningful geometric features of the local surface. It makes the reconstruction accuracy and completeness improved. Finally, a significant number of experimental results on real object data verify the superior performance of the proposed method. Our technique achieves the most delicate geometric details on thin objects for which the original depth map back-projection fusion scheme suffers from severe artifacts (See Fig.1).

Today's state-of-the-art methods for 3D object detection are based on lidar, stereo, or monocular cameras. Lidar-based methods achieve the best accuracy, but have a large footprint, high cost, and mechanically-limited angular sampling rates, resulting in low spatial resolution at long ranges. Recent approaches using low-cost monocular or stereo cameras promise to overcome these limitations but struggle in low-light or low-contrast regions as they rely on passive CMOS sensors. We propose a novel 3D object detection modality that exploits temporal illumination cues from a low-cost monocular gated imager. We introduce a novel deep detection architecture, Gated3D, that is tailored to temporal illumination cues in gated images. This modality allows us to exploit mature 2D object feature extractors that guide the 3D predictions through a frustum segment estimation. We assess the proposed method experimentally on a 3D detection dataset that includes gated images captured over 10,000 km of driving data. We validate that our method outperforms state-of-the-art monocular and stereo methods, opening up a new sensor modality as an avenue to replace lidar in autonomous driving. https://light.princeton.edu/gated3d

This paper tackles the task of category-level pose estimation for garments. With a near infinite degree of freedom, a garment's full configuration (i.e., poses) is often described by the per-vertex 3D locations of its entire 3D surface. However, garments are also commonly subject to extreme cases of self-occlusion, especially when folded or crumpled, making it challenging to perceive their full 3D surface. To address these challenges, we propose GarmentNets, where the key idea is to formulate the deformable object pose estimation problem as a shape completion task in the canonical space. This canonical space is defined across garments instances within a category, therefore, specifies the shared category-level pose. By mapping the observed partial surface to the canonical space and completing it in this space, the output representation describes the garment's full configuration using a complete 3D mesh with the per-vertex canonical coordinate label. To properly handle the thin 3D structure presented on garments, we proposed a novel 3D shape representation using the generalized winding number field. Experiments demonstrate that GarmentNets is able to generalize to unseen garment instances and achieve significantly better performance compared to alternative approaches. Code and data will be available online.

Gait recognition is one of the most important biometric technologies and has been applied in many fields. Recent gait recognition frameworks represent each gait frame by descriptors extracted from either global appearances or local regions of humans. However, the representations based on global information often neglect the details of the gait frame, while local region based descriptors cannot capture the relations among neighboring regions, thus reducing their discriminativeness. In this paper, we propose a novel feature extraction and fusion framework to achieve discriminative feature representations for gait recognition. Towards this goal, we take advantage of both global visual information and local region details and develop a Global and Local Feature Extractor (GLFE). Specifically, our GLFE module is composed of our newly designed multiple global and local convolutional layers (GLConv) to ensemble global and local features in a principle manner. Furthermore, we present a novel operation, namely Local Temporal Aggregation (LTA), to further preserve the spatial information by reducing the temporal resolution to obtain higher spatial resolution. With the help of our GLFE and LTA, our method significantly improves the discriminativeness of our visual features, thus improving the gait recognition performance. Extensive experiments demonstrate that our proposed method outperforms state-of-the-art gait recognition methods on two popular datasets.

Gait benchmarks empower the research community to train and evaluate high-performance gait recognition systems. Even though growing efforts have been devoted to cross-view recognition, academia is restricted by current existing databases captured in the controlled environment. In this paper, we contribute a new benchmark for Gait REcognition in the Wild (GREW). The GREW dataset is constructed from natural videos, which contains hundreds of cameras and thousands of hours streams in open systems. With tremendous manual annotations, the GREW consists of 26K identities and 128K sequences with rich attributes for unconstrained gait recognition. Moreover, we add a distractor set of over 233K sequences, making it more suitable for real-world applications. Compared with prevailing predefined cross-view datasets, the GREW has diverse and practical view variations, as well as more natural challenging factors. To the best of our knowledge, this is the first large-scale dataset for gait recognition in the wild. Equipped with this benchmark, we dissect the unconstrained gait recognition problem. Representative appearance-based and model-based methods are explored, and comprehensive baselines are established. Experimental results show (1) The proposed GREW benchmark is necessary for training and evaluating gait recognizer in the wild. (2) For state-of-the-art gait recognition approaches, there is a lot of room for improvement. (3) The GREW benchmark can be used as effective pre-training for controlled gait recognition. Benchmark website is https://www.grew-benchmark.org/.

We present GTT-Net, a supervised learning framework for the reconstruction of sparse dynamic 3D geometry. We build on a graph-theoretic formulation of the generalized trajectory triangulation problem, where non-concurrent multi-view imaging geometry is known but global image sequencing is not provided. GTT-Net learns pairwise affinities modeling the spatio-temporal relationships among our input observations and leverages them to determine 3D geometry estimates. Experiments reconstructing 3D motion-capture sequences show GTT-Net outperforms the state of the art in terms of accuracy and robustness. Within the context of articulated motion reconstruction, our proposed architecture is 1) able to learn and enforce semantic 3D motion priors for shared training and test domains, while being 2) able to generalize its performance across different training and test domains. Moreover, GTT-Net provides a computationally streamlined framework for trajectory triangulation with applications to multi-instance reconstruction and event segmentation.

We present a simple yet powerful neural network that implicitly represents and renders 3D objects and scenes only from 2D observations. The network models 3D geometries as a general radiance field, which takes a set of 2D images with camera poses and intrinsics as input, constructs an internal representation for each point of the 3D space, and then renders the corresponding appearance and geometry of that point viewed from an arbitrary position. The key to our approach is to learn local features for each pixel in 2D images and to then project these features to 3D points, thus yielding general and rich point representations. We additionally integrate an attention mechanism to aggregate pixel features from multiple 2D views, such that visual occlusions are implicitly taken into account. Extensive experiments demonstrate that our method can generate high-quality and realistic novel views for novel objects, unseen categories and challenging real-world scenes.

Panoptic segmentation as an integrated task of both static environmental understanding and dynamic object identification, has recently begun to receive broad research interest. In this paper, we propose a new computationally efficient LiDAR based panoptic segmentation framework, called GP-S3Net. GP-S3Net is a proposal-free approach in which no object proposals are needed to identify the objects in contrast to conventional two-stage panoptic systems, where a detection network is incorporated for capturing instance information. Our new design consists of a novel instance-level network to process the semantic results by constructing a graph convolutional network to identify objects (foreground), which later on are fused with the background classes. Through the fine-grained clusters of the foreground objects from the semantic segmentation backbone, over-segmentation priors are generated and subsequently processed by 3D sparse convolution to embed each cluster. Each cluster is treated as a node in the graph and its corresponding embedding is used as its node feature. Then a GCNN predicts whether edges exist between each cluster pair. We utilize the instance label to generate ground truth edge labels for each constructed graph in order to supervise the learning. Extensive experiments demonstrate that GP-S3Net outperforms the current state-of-the-art approaches, by a significant margin across available datasets such as, nuScenes and SemanticPOSS, ranking first on the competitive public SemanticKITTI leaderboard upon publication.

We introduce GNeRF, a framework to marry Generative Adversarial Networks (GAN) with Neural Radiance Field (NeRF) reconstruction for the complex scenarios with unknown and even randomly initialized camera poses. Recent NeRF-based advances have gained popularity for remarkable realistic novel view synthesis. However, most of them heavily rely on accurate camera poses estimation, while few recent methods can only optimize the unknown camera poses in roughly forward-facing scenes with relatively short camera trajectories and require rough camera poses initialization. Differently, our GNeRF only utilizes randomly initialized poses for complex outside-in scenarios. We propose a novel two-phases end-to-end framework. The first phase takes the use of GANs into the new realm for optimizing coarse camera poses and radiance fields jointly, while the second phase refines them with additional photometric loss. We overcome local minima using a hybrid and iterative optimization scheme. Extensive experiments on a variety of synthetic and natural scenes demonstrate the effectiveness of GNeRF. More impressively, our approach outperforms the baselines favorably in those scenes with repeated patterns or even low textures that are regarded as extremely challenging before.

In recent years, the growing number of medical imaging studies is placing an ever-increasing burden on radiologists. Deep learning provides a promising solution for automatic medical image analysis and clinical decision support. However, large-scale manually labeled datasets required for training deep neural networks are difficult and expensive to obtain for medical images. The purpose of this work is to develop label-efficient multimodal medical imaging representations by leveraging radiology reports. Specifically, we propose an attention-based framework (GLoRIA) for learning global and local representations by contrasting image sub-regions and words in the paired report. In addition, we propose methods to leverage the learned representations for various downstream medical image recognition tasks with limited labels. Our results demonstrate high-performance and label-efficiency for image-text retrieval, classification (finetuning and zeros-shot settings), and segmentation on different datasets.

We introduce the first Neural Architecture Search (NAS) method to find a better transformer architecture for image recognition. Recently, transformers without CNN-based backbones are found to achieve impressive performance for image recognition. However, the transformer is designed for NLP tasks and thus could be sub-optimal when directly used for image recognition. In order to improve the visual representation ability for transformers, we propose a new search space and searching algorithm. Specifically, we introduce a locality module that models the local correlations in images explicitly with fewer computational cost. With the locality module, our search space is defined to let the search algorithm freely trade off between global and local information as well as optimizing the low-level design choice in each module. To tackle the problem caused by huge search space, a hierarchical neural architecture search method is proposed to search the optimal vision transformer from two levels separately with the evolutionary algorithm. Extensive experiments on the ImageNet dataset demonstrate that our method can find more discriminative and efficient transformer variants than the ResNet family (e.g., ResNet101) and the baseline ViT for image classification.

Model compression techniques are recently gaining explosive attention for obtaining efficient AI models for various real time applications. Channel pruning is one important compression strategy, and widely used in slimming various DNNs. Previous gate-based or importance-based pruning methods aim to remove channels whose "importance" are smallest. However, it remains unclear what criteria the channel importance should be measured on, leading to various channel selection heuristics. Some other sampling-based pruning methods deploy sampling strategy to train sub-nets, which often causes the training instability and the compressed model's degraded performance. In view of the research gaps, we present a new module named Gates with Differentiable Polarization (GDP), inspired by principled optimization ideas. GDP can be plugged before convolutional layers without bells and whistles, to control the on-and-off of each channel or whole layer block. During the training process, the polarization effect will drive a subset of gates to smoothly decrease to exactly zero, while other gates gradually stay away from zero by a large margin. When training terminates, those zero-gated channels can be painlessly removed, while other non-zero gates can be absorbed into the succeeding convolution kernel, causing completely no interruption to training nor damage to the trained model. Experiments conducted over CIFAR-10 and ImageNet datasets show that the proposed GDP algorithm achieves the state-of-the-art performance on various benchmark DNNs at a broad range of pruning ratios. We also apply GDP to DeepLabV3Plus-ResNet50 on the challenging Pascal VOC segmentation task, whose test performance sees no drop (even slightly improved) with over 60% FLOPs saving.

We present GANcraft, an unsupervised neural rendering framework for generating photorealistic images of large 3D block worlds such as those created in Minecraft. Our method takes a semantic block world as input, where each block is assigned a semantic label such as dirt, grass, or water. We represent the world as a continuous volumetric function and train our model to render view-consistent photorealistic images for a user-controlled camera. In the absence of paired ground truth real images for the block world, we devise a training technique based on pseudo-ground truth and adversarial training. This stands in contrast to prior work on neural rendering for view synthesis, which requires ground truth images to estimate scene geometry and view-dependent appearance. In addition to camera trajectory, GANcraft allows user control over both scene semantics and output style. Experimental results with comparison to strong baselines show the effectiveness of GANcraft on this novel task of photorealistic 3D block world synthesis.

We present a framework for training GANs with explicit control over generated facial images. We are able to control the generated image by settings exact attributes such as age, pose, expression, etc. Most approaches for manipulating GAN-generated images achieve partial control by leveraging the latent space disentanglement properties, obtained implicitly after standard GAN training. Such methods are able to change the relative intensity of certain attributes, but not explicitly set their values. Recently proposed methods, designed for explicit control over human faces, harness morphable 3D face models (3DMM) to allow fine-grained control capabilities in GANs. Unlike these methods, our control is not constrained to 3DMM parameters and is extendable beyond the domain of human faces. Using contrastive learning, we obtain GANs with an explicitly disentangled latent space. This disentanglement is utilized to train control-encoders mapping human-interpretable inputs to suitable latent vectors, thus allowing explicit control. In the domain of human faces we demonstrate control over identity, age, pose, expression, hair color and illumination. We also demonstrate control capabilities of our framework in the domains of painted portraits and dog image generation. We demonstrate that our approach achieves state-of-the-art performance both qualitatively and quantitatively.

For successful semantic editing of real images, it is critical for a GAN inversion method to find an in-domain latent code that aligns with the domain of a pre-trained GAN model. Unfortunately, such in-domain latent codes can be found only for in-range images that align with the training images of a GAN model. In this paper, we propose BDInvert, a novel GAN inversion approach to semantic editing of out-of-range images that are geometrically unaligned with the training images of a GAN model. To find a latent code that is semantically editable, BDInvert inverts an input out-of-range image into an alternative latent space than the original latent space. We also propose a regularized inversion method to find a solution that supports semantic editing in the alternative space. Our experiments show that BDInvert effectively supports semantic editing of out-of-range images with geometric transformations.

In this paper, we investigate the knowledge distillation (KD) strategy for object detection and propose an effective framework applicable to both homogeneous and heterogeneous student-teacher pairs. The conventional feature imitation paradigm introduces imitation masks to focus on informative foreground areas while excluding the background noises. However, we find that those methods fail to fully utilize the semantic information in all feature pyramid levels, which leads to inefficiency for knowledge distillation between FPN-based detectors. To this end, we propose a novel semantic-guided feature imitation technique, which automatically performs soft matching between feature pairs across all pyramid levels to provide the optimal guidance to the student. To push the envelop even further, we introduce contrastive distillation to effectively capture the information encoded in the relationship between different feature regions. Finally, we propose a generalized detection KD pipeline, which is capable of distilling both homogeneous and heterogeneous detector pairs. Our method consistently outperforms the existing detection KD techniques, and works when (1) components in the framework are used separately and in conjunction; (2) for both homogeneous and heterogenous student-teacher pairs and (3) on multiple detection benchmarks. With a powerful X101-FasterRCNN-Instaboost detector as the teacher, R50-FasterRCNN reaches 44.0% AP, R50-RetinaNet reaches 43.3% AP and R50-FCOS reaches 43.1% AP on COCO dataset.

We contribute to approximate algorithms for the quadratic assignment problem also known as graph matching. Inspired by the success of the fusion moves technique developed for multilabel discrete Markov random fields, we investigate its applicability to graph matching. In particular, we show how fusion moves can be efficiently combined with the dedicated state-of-the-art dual methods that have recently shown superior results in computer vision and bio-imaging applications. As our empirical evaluation on a wide variety of graph matching datasets suggests, fusion moves significantly improve performance of these methods in terms of speed and quality of the obtained solutions. Our method sets a new state-of-the-art with a notable margin with respect to its competitors.

Transformer, as a strong and flexible architecture for modelling long-range relations, has been widely explored in vision tasks. However, when used in video inpainting that requires fine-grained representation, existed method still suffers from yielding blurry edges in detail due to the hard patch splitting. Here we aim to tackle this problem by proposing FuseFormer, a Transformer model designed for video inpainting via fine-grained feature fusion based on novel Soft Split and Soft Composition operations. The soft split divides feature map into many patches with given overlapping interval. On the contrary, the soft composition operates by stitching different patches into a whole feature map where pixels in overlapping regions are summed up. These two modules are first used in tokenization before Transformer layers and de-tokenization after Transformer layers, for effective mapping between tokens and features. Therefore, sub-patch level information interaction is enabled for more effective feature propagation between neighboring patches, resulting in synthesizing vivid content for hole regions in videos. Moreover, in FuseFormer, we elaborately insert the soft composition and soft split into the feed-forward network, enabling the 1D linear layers to have the capability of modelling 2D structure. And, the sub-patch level feature fusion ability is further enhanced. In both quantitative and qualitative evaluations, our proposed FuseFormer surpasses state-of-the-art methods. We also conduct detailed analysis to examine its superiority.