CVPR 2022 论文列表

We propose a novel one-step transformer-based person search framework, PSTR, that jointly performs person detection and re-identification (re-id) in a single architecture. PSTR comprises a person search-specialized (PSS) module that contains a detection encoder-decoder for person detection along with a discriminative re-id decoder for person re-id. The discriminative re-id decoder utilizes a multi-level supervision scheme with a shared decoder for discriminative re-id feature learning and also comprises a part attention block to encode relationship between different parts of a person. We further introduce a simple multi-scale scheme to support re-id across person instances at different scales. PSTR jointly achieves the diverse objectives of object-level recognition (detection) and instance-level matching (re-id). To the best of our knowledge, we are the first to propose an end-to-end one-step transformer-based person search framework. Experiments are performed on two popular benchmarks: CUHK-SYSU and PRW. Our extensive ablations reveal the merits of the proposed contributions. Further, the proposed PSTR sets a new state-of-the-art on both benchmarks. On the challenging PRW benchmark, PSTR achieves a mean average precision (mAP) score of 56.5%. The source code is available at https://github.com/JialeCao001/PSTR.

In this paper, we propose a new deep learning-based method for estimating room layout given a pair of 360 panoramas. Our system, called Position-aware Stereo Merging Network or PSMNet, is an end-to-end joint layout-pose estimator. PSMNet consists of a Stereo Pano Pose (SP^2) transformer and a novel Cross-Perspective Projection (CP^2) layer. The stereo-view SP^2 transformer is used to implicitly infer correspondences between views, and can handle noisy poses. The pose-aware CP^2layer is designed to render features from the adjacent view to the anchor (reference) view, in order to perform view fusion and estimate the visible layout. Our experiments and analysis validate our method, which significantly outperforms the state-of-the-art layout estimators, especially for large and complex room spaces.

Scene Graph Generation (SGG) has attracted more and more attention from visual researchers in recent years, since Scene Graph (SG) is valuable in many downstream tasks due to its rich structural-semantic details. However, the application value of SG on downstream tasks is severely limited by the predicate classification bias, which is caused by long-tailed data and presented as semantic bias of predicted relation predicates. Existing methods mainly reduce the prediction bias by better aggregating contexts and integrating external priori knowledge, but rarely take the semantic similarities between predicates into account. In this paper, we propose a Predicate Probability Distribution based Loss (PPDL) to train the biased SGG models and obtain unbiased Scene Graphs ultimately. Firstly, we propose a predicate probability distribution as the semantic representation of a particular predicate class. Afterwards, we re-balance the biased training loss according to the similarity between the predicted probability distribution and the estimated one, and eventually eliminate the long-tailed bias on predicate classification. Notably, the PPDL training method is model-agnostic, and extensive experiments and qualitative analyses on the Visual Genome dataset reveal significant performance improvements of our method on tail classes compared to the state-of-the-art methods.

State-of-the-art approaches to ObjectGoal navigation (ObjectNav) rely on reinforcement learning and typically require significant computational resources and time for learning. We propose Potential functions for ObjectGoal Navigation with Interaction-free learning (PONI), a modular approach that disentangles the skills of 'where to look?' for an object and 'how to navigate to (x, y)?'. Our key insight is that 'where to look?' can be treated purely as a perception problem, and learned without environment interactions. To address this, we propose a network that predicts two complementary potential functions conditioned on a semantic map and uses them to decide where to look for an unseen object. We train the potential function network using supervised learning on a passive dataset of top-down semantic maps, and integrate it into a modular framework to perform ObjectNav. Experiments on Gibson and Matterport3D demonstrate that our method achieves the state-of-the-art for ObjectNav while incurring up to 1,600x less computational cost for training. Code and pre-trained models are available.

Implicit neural networks have been successfully used for surface reconstruction from point clouds. However, many of them face scalability issues as they encode the isosurface function of a whole object or scene into a single latent vector. To overcome this limitation, a few approaches infer latent vectors on a coarse regular 3D grid or on 3D patches, and interpolate them to answer occupancy queries. In doing so, they loose the direct connection with the input points sampled on the surface of objects, and they attach information uniformly in space rather than where it matters the most, i.e., near the surface. Besides, relying on fixed patch sizes may require discretization tuning. To address these issues, we propose to use point cloud convolutions and compute latent vectors at each input point. We then perform a learning-based interpolation on nearest neighbors using inferred weights. Experiments on both object and scene datasets show that our approach significantly outperforms other methods on most classical metrics, producing finer details and better reconstructing thinner volumes. The code is available at https://github.com/valeoai/POCO

Label noise has been a practical challenge in deep learning due to the strong capability of deep neural networks in fitting all training data. Prior literature primarily resorts to sample selection methods for combating noisy labels. However, these approaches focus on dividing samples by order sorting or threshold selection, inevitably introducing hyper-parameters (e.g., selection ratio / threshold) that are hard-to-tune and dataset-dependent. To this end, we propose a simple yet effective approach named PNP (Probabilistic Noise Prediction) to explicitly model label noise. Specifically, we simultaneously train two networks, in which one predicts the category label and the other predicts the noise type. By predicting label noise probabilistically, we identify noisy samples and adopt dedicated optimization objectives accordingly. Finally, we establish a joint loss for network update by unifying the classification loss, the auxiliary constraint loss, and the in-distribution consistency loss. Comprehensive experimental results on synthetic and real-world datasets demonstrate the superiority of our proposed method.

Inferring programs which generate 2D and 3D shapes is important for reverse engineering, editing, and more. Training models to perform this task is complicated because paired (shape, program) data is not readily available for many domains, making exact supervised learning infeasible. However, it is possible to get paired data by compromising the accuracy of either the assigned program labels or the shape distribution. Wake-sleep methods use samples from a generative model of shape programs to approximate the distribution of real shapes. In self-training, shapes are passed through a recognition model, which predicts programs that are treated as 'pseudo-labels' for those shapes. Related to these approaches, we introduce a novel self-training variant unique to program inference, where program pseudo-labels are paired with their executed output shapes, avoiding label mismatch at the cost of an approximate shape distribution. We propose to group these regimes under a single conceptual framework, where training is performed with maximum likelihood updates sourced from either Pseudo-Labels or an Approximate Distribution (PLAD). We evaluate these techniques on multiple 2D and 3D shape program inference domains. Compared with policy gradient reinforcement learning, we show that PLAD techniques infer more accurate shape programs and converge significantly faster. Finally, we propose to combine updates from different PLAD methods within the training of a single model, and find that this approach outperforms any individual technique.

We present a novel method to learn Personalized Implicit Neural Avatars (PINA) from a short RGB-D sequence. This allows non-expert users to create a detailed and personalized virtual copy of themselves, which can be animated with realistic clothing deformations. PINA does not require complete scans, nor does it require a prior learned from large datasets of clothed humans. Learning a complete avatar in this setting is challenging, since only few depth observations are available, which are noisy and incomplete (i.e. only partial visibility of the body per frame). We propose a method to learn the shape and non-rigid deformations via a pose-conditioned implicit surface and a deformation field, defined in canonical space. This allows us to fuse all partial observations into a single consistent canonical representation. Fusion is formulated as a global optimization problem over the pose, shape and skinning parameters. The method can learn neural avatars from real noisy RGB-D sequences for a diverse set of people and clothing styles and these avatars can be animated given unseen motion sequences.

Generative model based image lossless compression algorithms have seen a great success in improving compression ratio. However, the throughput for most of them is less than 1 MB/s even with the most advanced AI accelerated chips, preventing them from most real-world applications, which often require 100 MB/s. In this paper, we propose PILC, an end-to-end image lossless compression framework that achieves 200 MB/s for both compression and decompression with a single NVIDIA Tesla V100 GPU, 10x faster than the most efficient one before. To obtain this result, we first develop an AI codec that combines auto-regressive model and VQ-VAE which performs well in lightweight setting, then we design a low complexity entropy coder that works well with our codec. Experiments show that our framework compresses better than PNG by a margin of 30% in multiple datasets. We believe this is an important step to bring AI compression forward to commercial use.

Intrinsic image decomposition is the process of recovering the image formation components (reflectance and shading) from an image. Previous methods employ either explicit priors to constrain the problem or implicit constraints as formulated by their losses (deep learning). These methods can be negatively influenced by strong illumination conditions causing shading-reflectance leakages. Therefore, in this paper, an end-to-end edge-driven hybrid CNN approach is proposed for intrinsic image decomposition. Edges correspond to illumination invariant gradients. To handle hard negative illumination transitions, a hierarchical approach is taken including global and local refinement layers. We make use of attention layers to further strengthen the learning process. An extensive ablation study and large scale experiments are conducted showing that it is beneficial for edge-driven hybrid IID networks to make use of illumination invariant descriptors and that separating global and local cues helps in improving the performance of the network. Finally, it is shown that the proposed method obtains state of the art performance and is able to generalise well to real world images. The project page with pretrained models, finetuned models and network code can be found at: https://ivi.fnwi.uva.nl/cv/pienet/

Domain generalization refers to the problem of training a model from a collection of different source domains that can directly generalize to the unseen target domains. A promising solution is contrastive learning, which attempts to learn domain-invariant representations by exploiting rich semantic relations among sample-to-sample pairs from different domains. A simple approach is to pull positive sample pairs from different domains closer while pushing other negative pairs further apart. In this paper, we find that directly applying contrastive-based methods (e.g., supervised contrastive learning) are not effective in domain generalization. We argue that aligning positive sample-to-sample pairs tends to hinder the model generalization due to the significant distribution gaps between different domains. To address this issue, we propose a novel proxy-based contrastive learning method, which replaces the original sample-to-sample relations with proxy-to-sample relations, significantly alleviating the positive alignment issue. Experiments on the four standard benchmarks demonstrate the effectiveness of the proposed method. Furthermore, we also consider a more complex scenario where no ImageNet pre-trained models are provided. Our method consistently shows better performance.

Photorealistic style transfer entails transferring the style of a reference image to another image so the result seems like a plausible photo. Our work is inspired by the observation that existing models are slow due to their large sizes. We introduce PCA-based knowledge distillation to distill lightweight models and show it is motivated by theory. To our knowledge, this is the first knowledge distillation method for photorealistic style transfer. Our experiments demonstrate its versatility for use with different backbone architectures, VGG and MobileNet, across six image resolutions. Compared to existing models, our top-performing model runs at speeds 5-20x faster using at most 1% of the parameters. Additionally, our distilled models achieve a better balance between stylization strength and content preservation than existing models. To support reproducing our method and models, we share the code at https://github.com/chiutaiyin/PCA-Knowledge-Distillation.

In this paper, we study the problem of procedure planning in instructional videos. Here, an agent must produce a plausible sequence of actions that can transform the environment from a given start to a desired goal state. When learning procedure planning from instructional videos, most recent work leverages intermediate visual observations as supervision, which requires expensive annotation efforts to localize precisely all the instructional steps in training videos. In contrast, we remove the need for expensive temporal video annotations and propose a weakly supervised approach by learning from natural language instructions. Our model is based on a transformer equipped with a memory module, which maps the start and goal observations to a sequence of plausible actions. Furthermore, we augment our model with a probabilistic generative module to capture the uncertainty inherent to procedure planning, an aspect largely overlooked by previous work. We evaluate our model on three datasets and show our weakly-supervised approach outperforms previous fully supervised state-of-the-art models on multiple metrics.

Monocular depth estimation is vital for scene understanding and downstream tasks. We focus on the supervised setup, in which ground-truth depth is available only at training time. Based on knowledge about the high regularity of real 3D scenes, we propose a method that learns to selectively leverage information from coplanar pixels to improve the predicted depth. In particular, we introduce a piecewise planarity prior which states that for each pixel, there is a seed pixel which shares the same planar 3D surface with the former. Motivated by this prior, we design a network with two heads. The first head outputs pixel-level plane coefficients, while the second one outputs a dense offset vector field that identifies the positions of seed pixels. The plane coefficients of seed pixels are then used to predict depth at each position. The resulting prediction is adaptively fused with the initial prediction from the first head via a learned confidence to account for potential deviations from precise local planarity. The entire architecture is trained end-to-end thanks to the differentiability of the proposed modules and it learns to predict regular depth maps, with sharp edges at occlusion boundaries. An extensive evaluation of our method shows that we set the new state of the art in supervised monocular depth estimation, surpassing prior methods on NYU Depth-v2 and on the Garg split of KITTI. Our method delivers depth maps that yield plausible 3D reconstructions of the input scenes. Code is available at: https://github.com/SysCV/P3Depth

Traditional object detectors are ill-equipped for incremental learning. However, fine-tuning directly on a well-trained detection model with only new data will lead to catastrophic forgetting. Knowledge distillation is a flexible way to mitigate catastrophic forgetting. In Incremental Object Detection (IOD), previous work mainly focuses on distilling for the combination of features and responses. However, they under-explore the information that contains in responses. In this paper, we propose a response-based incremental distillation method, dubbed Elastic Response Distillation (ERD), which focuses on elastically learning responses from the classification head and the regression head. Firstly, our method transfers category knowledge while equipping student detector with the ability to retain localization information during incremental learning. In addition, we further evaluate the quality of all locations and provide valuable responses by the Elastic Response Selection (ERS) strategy. Finally, we elucidate that the knowledge from different responses should be assigned with different importance during incremental distillation. Extensive experiments conducted on MS COCO demonstrate the proposed method achieves state-of-the-art performance, which substantially narrows the performance gap towards full training.

In real-world applications, it is important and desirable to learn a model that performs well on out-of-distribution (OOD) data. Recently, causality has become a powerful tool to tackle the OOD generalization problem, with the idea resting on the causal mechanism that is invariant across domains of interest. To leverage the generally unknown causal mechanism, existing works assume a linear form of causal feature or require sufficiently many and diverse training domains, which are usually restrictive in practice. In this work, we obviate these assumptions and tackle the OOD problem without explicitly recovering the causal feature. Our approach is based on transformations that modify the non-causal feature but leave the causal part unchanged, which can be either obtained from prior knowledge or learned from the training data in the multi-domain scenario. Under the setting of invariant causal mechanism, we theoretically show that if all such transformations are available, then we can learn a minimax optimal model across the domains using only single domain data. Noticing that knowing a complete set of these causal invariant transformations may be impractical, we further show that it suffices to know only a subset of these transformations. Based on the theoretical findings, a regularized training procedure is proposed to improve the OOD generalization capability. Extensive experimental results on both synthetic and real datasets verify the effectiveness of the proposed algorithm, even with only a few causal invariant transformations.

This paper proposes a new eXplanation framework, called OrphicX, for generating causal explanations for any graph neural networks (GNNs) based on learned latent causal factors. Specifically, we construct a distinct generative model and design an objective function that encourages the generative model to produce causal, compact, and faithful explanations. This is achieved by isolating the causal factors in the latent space of graphs by maximizing the information flow measurements. We theoretically analyze the cause-effect relationships in the proposed causal graph, identify node attributes as confounders between graphs and GNN predictions, and circumvent such confounder effect by leveraging the backdoor adjustment formula. Our framework is compatible with any GNNs, and it does not require access to the process by which the target GNN produces its predictions. In addition, it does not rely on the linear-independence assumption of the explained features, nor require prior knowledge on the graph learning tasks. We show a proof-of-concept of OrphicX on canonical classification problems on graph data. In particular, we analyze the explanatory subgraphs obtained from explanations for molecular graphs (i.e., Mutag) and quantitatively evaluate the explanation performance with frequently occurring subgraph patterns. Empirically, we show that OrphicX can effectively identify the causal semantics for generating causal explanations, significantly outperforming its alternatives.

In contrast to the generic object, aerial targets are often non-axis aligned with arbitrary orientations having the cluttered surroundings. Unlike the mainstreamed approaches regressing the bounding box orientations, this paper proposes an effective adaptive points learning approach to aerial object detection by taking advantage of the adaptive points representation, which is able to capture the geometric information of the arbitrary-oriented instances. To this end, three oriented conversion functions are presented to facilitate the classification and localization with accurate orientation. Moreover, we propose an effective quality assessment and sample assignment scheme for adaptive points learning toward choosing the representative oriented reppoints samples during training, which is able to capture the non-axis aligned features from adjacent objects or background noises. A spatial constraint is introduced to penalize the outlier points for roust adaptive learning. Experimental results on four challenging aerial datasets including DOTA, HRSC2016, UCAS-AOD and DIOR-R, demonstrate the efficacy of our proposed approach. The source code is availabel at: https://github.com/LiWentomng/OrientedRepPoints.

Video prediction is an extrapolation task that predicts future frames given past frames, and video frame interpolation is an interpolation task that estimates intermediate frames between two frames. We have witnessed the tremendous advancement of video frame interpolation, but the general video prediction in the wild is still an open question. Inspired by the photo-realistic results of video frame interpolation, we present a new optimization framework for video prediction via video frame interpolation, in which we solve an extrapolation problem based on an interpolation model. Our video prediction framework is based on optimization with a pretrained differentiable video frame interpolation module without the need for a training dataset, and thus there is no domain gap issue between training and test data. Also, our approach does not need any additional information such as semantic or instance maps, which makes our framework applicable to any video. Extensive experiments on the Cityscapes, KITTI, DAVIS, Middlebury, and Vimeo90K datasets show that our video prediction results are robust in general scenarios, and our approach outperforms other video prediction methods that require a large amount of training data or extra semantic information.