ICCV 2021 论文列表

In semantic segmentation tasks, input images can often have more than one plausible interpretation, thus allowing for multiple valid labels. To capture such ambiguities, recent work has explored the use of probabilistic networks that can learn a distribution over predictions. However, these do not necessarily represent the empirical distribution accurately. In this work, we present a strategy for learning a calibrated predictive distribution over semantic maps, where the probability associated with each prediction reflects its ground truth correctness likelihood. To this end, we propose a novel two-stage, cascaded approach for calibrated adversarial refinement: (i) a standard segmentation network is trained with categorical cross-entropy to predict a pixelwise probability distribution over semantic classes and (ii) an adversarially trained stochastic network is used to model the inter-pixel correlations to refine the output of the first network into coherent samples. Importantly, to calibrate the refinement network and prevent mode collapse, the expectation of the samples in the second stage is matched to the probabilities predicted in the first. We demonstrate the versatility and robustness of the approach by achieving state-of-the-art results on the multigrader LIDC dataset and on a modified Cityscapes dataset with injected ambiguities. In addition, we show that the core design can be adapted to other tasks requiring learning a calibrated predictive distribution by experimenting on a toy regression dataset. We provide an open source implementation of our method at https://github.com/EliasKassapis/CARSSS.

A large gap exists between fully-supervised object detection and weakly-supervised object detection. To narrow this gap, some methods consider knowledge transfer from additional fully-supervised dataset. But these methods do not fully exploit discriminative category information in the fully-supervised dataset, thus causing low mAP. To solve this issue, we propose a novel category transfer framework for weakly supervised object detection. The intuition is to fully leverage both visually-discriminative and semantically-correlated category information in the fully-supervised dataset to enhance the object-classification ability of a weakly-supervised detector. To handle overlapping category transfer, we propose a double-supervision mean teacher to gather common category information and bridge the domain gap between two datasets. To handle non-overlapping category transfer, we propose a semantic graph convolutional network to promote the aggregation of semantic features between correlated categories. Experiments are conducted with Pascal VOC 2007 as the target weakly-supervised dataset and COCO as the source fully-supervised dataset. Our category transfer framework achieves 63.5% mAP and 80.3% CorLoc with 5 overlapping categories between two datasets, which outperforms the state-of-the-art methods. Codes are avaliable at https://github.com/MediaBrain-SJTU/CaT.

Single image camera calibration is the task of estimating the camera parameters from a single input image, such as the vanishing points, focal length, and horizon line. In this work, we propose Camera calibration TRansformer with Line-Classification (CTRL-C), an end-to-end neural network-based approach to single image camera calibration, which directly estimates the camera parameters from an image and a set of line segments. Our network adopts the transformer architecture to capture the global structure of an image with multi-modal inputs in an end-to-end manner. We also propose an auxiliary task of line classification to train the network to extract the global geometric information from lines effectively. Our experiments demonstrate that CTRL-C outperforms the previous state-of-the-art methods on the Google Street View and SUN360 benchmark datasets. Code is available at https://github.com/jwlee-vcl/CTRL-C.

Generating an interpretable and compact representation of 3D shapes from point clouds is an important and challenging problem. This paper presents CSG-Stump Net, an unsupervised end-to-end network for learning shapes from point clouds and discovering the underlying constituent modeling primitives and operations as well. At the core is a three-level structure called CSG-Stump , consisting of a complement layer at the bottom, an intersection layer in the middle, and a union layer at the top. CSG-Stump is proven to be equivalent to CSG in terms of representation, therefore inheriting the interpretable, compact and editable nature of CSG while freeing from CSG's complex tree structures. Particularly, the CSG-Stump has a simple and regular structure, allowing neural networks to give outputs of a constant dimensionality, which makes itself deep-learning friendly. Due to these characteristics of CSG-Stump, CSG-Stump Net achieves superior results compared to previous CSG-based methods and generates much more appealing shapes, as confirmed by extensive experiment

Recent deep generative inpainting methods use attention layers to allow the generator to explicitly borrow feature patches from the known region to complete a missing region. Due to the lack of supervision signals for the correspondence between missing regions and known regions, it may fail to find proper reference features, which often leads to artifacts in the results. Also, it computes pair-wise similarity across the entire feature map during inference bringing a significant computational overhead. To address this issue, we propose to teach such patch-borrowing behavior to an attention-free generator by joint training of an auxiliary contextual reconstruction task, which encourages the generated output to be plausible even when reconstructed by surrounding regions. The auxiliary branch can be seen as a learnable loss function, i.e. named as contextual reconstruction (CR) loss, where query-reference feature similarity and reference-based reconstructor are jointly optimized with the inpainting generator. The auxiliary branch (i.e. CR loss) is required only during training, and only the inpainting generator is required during the inference. Experimental results demonstrate that the proposed inpainting model compares favourably against the state-of-the-art in terms of quantitative and visual performance. Code is available at https://github.com/zengxianyu/crfill.

Representing human-made objects as a collection of base primitives has a long history in computer vision and reverse engineering. In the case of high-resolution point cloud scans, the challenge is to be able to detect both large primitives as well as those explaining the detailed parts. While the classical RANSAC approach requires case-specific parameter tuning, state-of-the-art networks are limited by memory consumption of their backbone modules such as PointNet++, and hence fail to detect the fine-scale primitives. We present Cascaded Primitive Fitting Networks (CPFN) that relies on an adaptive patch sampling network to assemble detection results of global and local primitive detection networks. As a key enabler, we present a merging formulation that dynamically aggregates the primitives across global and local scales. Our evaluation demonstrates that CPFN improves the state-of-the-art SPFN performance by 13-14% on high-resolution point cloud datasets and specifically improves the detection of fine-scale primitives by 20-22%. Our code is available at: https://github.com/erictuanle/CPFN

Modeling the hand-object (HO) interaction not only requires estimation of the HO pose, but also pays attention to the contact due to their interaction. Significant progress has been made in estimating hand and object separately with deep learning methods, simultaneous HO pose estimation and contact modeling has not yet been fully explored. In this paper, we present an explicit contact representation namely Contact Potential Field (CPF), and a learning-fitting hybrid framework namely MIHO to Modeling the Interaction of Hand and Object. In CPF, we treat each contacting HO vertex pair as a spring-mass system. Hence the whole system forms a potential field with minimal elastic energy at the grasp position. Extensive experiments on the two commonly used benchmarks have demonstrated that our method can achieve state-of-the-art in several reconstruction metrics, and allow us to produce more physically plausible HO pose even when the ground-truth exhibits severe interpenetration or disjointedness. Our code is available at https://github.com/lixiny/CPF.

We propose a novel framework for finding correspondences in images based on a deep neural network that, given two images and a query point in one of them, finds its correspondence in the other. By doing so, one has the option to query only the points of interest and retrieve sparse correspondences, or to query all points in an image and obtain dense mappings. Importantly, in order to capture both local and global priors, and to let our model relate between image regions using the most relevant among said priors, we realize our network using a transformer. At inference time, we apply our correspondence network by recursively zooming in around the estimates, yielding a multi-scale pipeline able to provide highly-accurate correspondences. Our method significantly outperforms the state-of-the-art on both sparse and dense correspondence problems on multiple datasets and tasks, ranging from wide-baseline stereo to optical flow, without any retraining for a specific dataset.

There has been a recent surge of interest in cross-modal pre-training. However, existed approaches pre-train a one-stream model to learn joint vision-language representation, which suffers from calculation explosion when conducting cross-modal retrieval. In this work, we propose the Contrastive Cross-Modal Knowledge Sharing Pre-training (COOKIE) method to learn universal text-image representations. There are two key designs in it, one is the weight-sharing transformer on top of the visual and textual encoders to align text and image semantically, the other is three kinds of contrastive learning designed for sharing knowledge between different modalities. Cross-modal knowledge sharing greatly promotes the learning of unimodal representation. Experiments on multi-modal matching tasks including cross-modal retrieval, text matching, and image retrieval show the effectiveness and efficiency of our pre-training framework. Our COOKIE fine-tuned on cross-modal datasets MSCOCO, Flickr30K, and MSRVTT achieves new state-of-the-art results while using only 3/1000 inference time comparing to one-stream models. There are also 5.7 and 3.9 improvements in the task of image retrieval and text matching. Source code will be made public.

Most video super-resolution methods focus on restoring high-resolution video frames from low-resolution videos without taking into account compression. However, most videos on the web or mobile devices are compressed, and the compression can be severe when the bandwidth is limited. In this paper, we propose a new compression-informed video super-resolution model to restore high-resolution content without introducing artifacts caused by compression. The proposed model consists of three modules for video super-resolution: bi-directional recurrent warping, detail-preserving flow estimation, and Laplacian enhancement. All these three modules are used to deal with compression properties such as the location of the intra-frames in the input and smoothness in the output frames. For thorough performance evaluation, we conducted extensive experiments on standard datasets with a wide range of compression rates, covering many real video use cases. We showed that our method not only recovers high-resolution content on uncompressed frames from the widely-used benchmark datasets, but also achieves state-of-the-art performance in super-resolving compressed videos based on numerous quantitative metrics. We also evaluated the proposed method by simulating streaming from YouTube to demonstrate its effectiveness and robustness. The source codes and trained models are available at https://github.com/google-research/google-research/tree/master/comisr.

Overconfident predictions on out-of-distribution (OOD) samples is a thorny issue for deep neural networks. The key to resolve the OOD overconfidence issue inherently is to build a subset of OOD samples and then suppress predictions on them. This paper proposes the Chamfer OOD examples (CODEs), whose distribution is close to that of in-distribution samples, and thus could be utilized to alleviate the OOD overconfidence issue effectively by suppressing predictions on them. To obtain CODEs, we first generate seed OOD examples via slicing&splicing operations on in-distribution samples from different categories, and then feed them to the Chamfer generative adversarial network for distribution transformation, without accessing to any extra data. Training with suppressing predictions on CODEs is validated to alleviate the OOD overconfidence issue largely without hurting classification accuracy, and outperform the state-of-the-art methods. Besides, we demonstrate CODEs are useful for improving OOD detection and classification.

Visible-Infrared person re-identification (VI-ReID) aims to match cross-modality pedestrian images, breaking through the limitation of single-modality person ReID in dark environment. In order to mitigate the impact of large modality discrepancy, existing works manually design various two-stream architectures to separately learn modality-specific and modality-sharable representations. Such a manual design routine, however, highly depends on massive experiments and empirical practice, which is time consuming and labor intensive. In this paper, we systematically study the manually designed architectures, and identify that appropriately separating Batch Normalization (BN) layers is the key to bring a great boost towards cross-modality matching. Based on this observation, the essential objective is to find the optimal separation scheme for each BN layer. To this end, we propose a novel method, named Cross-Modality Neural Architecture Search (CM-NAS). It consists of a BN-oriented search space in which the standard optimization can be fulfilled subject to the cross-modality task. Equipped with the searched architecture, our method outperforms state-of-the-art counterparts in both two benchmarks, improving the Rank-1/mAP by 6.70%/6.13% on SYSU-MM01 and by 12.17%/11.23% on RegDB. Code is released at https://github.com/JDAI-CV/CM-NAS.

The data poisoning attack has raised serious security concerns on the safety of deep neural networks since it can lead to neural backdoor that misclassifies certain inputs crafted by an attacker. In particular, the sample-targeted backdoor attack is a new challenge. It targets at one or a few specific samples, called target samples, to misclassify them to a target class. Without a trigger planted in the backdoor model, the existing backdoor detection schemes fail to detect the sample-targeted backdoor as they depend on reverse-engineering the trigger or strong features of the trigger. In this paper, we propose a novel scheme to detect and mitigate sample-targeted backdoor attacks. We discover and demonstrate a unique property of the sample-targeted backdoor, which forces a boundary change such that small "pockets" are formed around the target sample. Based on this observation, we propose a novel defense mechanism to pinpoint a malicious pocket by "wrapping" them into a tight convex hull in the feature space. We design an effective algorithm to search for such a convex hull and remove the backdoor by fine-tuning the model using the identified malicious samples with the corrected label according to the convex hull. The experiments show that the proposed approach is highly efficient for detecting and mitigating a wide range of sample-targeted backdoor attacks.

We present a two-stage pre-training approach that improves the generalization ability of standard single-domain pre-training. While standard pre-training on a single large dataset (such as ImageNet) can provide a good initial representation for transfer learning tasks, this approach may result in biased representations that impact the success of learning with new multi-domain data (e.g., different artistic styles) via methods like domain adaptation. We propose a novel pre-training approach called Cross-Domain Self-supervision (CDS), which directly employs unlabeled multi-domain data for downstream domain transfer tasks. Our approach uses self-supervision not only within a single domain but also across domains. In-domain instance discrimination is used to learn discriminative features on new data in a domain-adaptive manner, while cross-domain matching is used to learn domain-invariant features. We apply our method as a second pre-training step (after ImageNet pre-training), resulting in a significant target accuracy boost to diverse domain transfer tasks compared to standard one-stage pre-training.

Nuclear instance segmentation is a challenging task due to a large number of touching and overlapping nuclei in pathological images. Existing methods cannot effectively recognize the accurate boundary owing to neglecting the relationship between pixels (e.g., direction information). In this paper, we propose a novel Centripetal Direction Network (CDNet) for nuclear instance segmentation. Specifically, we define the centripetal direction feature as a class of adjacent directions pointing to the nuclear center to represent the spatial relationship between pixels within the nucleus. These direction features are then used to construct a direction difference map to represent the similarity within instances and the differences between instances. Finally, we propose a direction-guided refinement module, which acts as a plug-and-play module to effectively integrate auxiliary tasks and aggregate the features of different branches. Experiments on MoNuSeg and CPM17 datasets show that CDNet is significantly better than the other methods and achieves the state-of-the-art performance. The code is available at https://github.com/honglianghe/CDNet.

A medical imaging model is usually explored for the diagnosis of a single disease. However, with the expanding demand for multi-disease diagnosis in clinical applications, multi-function solutions need to be investigated. Previous works proposed to either exploit different disease labels to conduct transfer learning through fine-tuning, or transfer knowledge across different domains with similar diseases. However, these methods still cannot address the real clinical challenge - a multi-disease model is required but annotations for each disease are not always available. In this paper, we introduce the task of transferring knowledge from single-disease diagnosis (source domain) to enhance multi-disease diagnosis (target domain). A category-invariant cross-domain transfer (CCT) method is proposed to address this single-to-multiple extension. First, for domain-specific task learning, we present a confidence weighted pooling (CWP) to obtain coarse heatmaps for different disease categories. Then, conditioned on these heatmaps, category-invariant feature refinement (CIFR) blocks are proposed to better localize discriminative semantic regions related to the corresponding diseases. The category-invariant characteristic enables transferability from the source domain to the target domain. We validate our method in two popular areas: extending diabetic retinopathy to identifying multiple ocular diseases, and extending glioma identification to the diagnosis of other brain tumors.

In this work, we tackle the problem of category-level online pose tracking for objects from point cloud sequences. For the first time, we propose a unified framework that can handle 9DoF object pose tracking for novel rigid object instances as well as per-part pose tracking for articulated objects from known categories. Here the 9DoF pose, comprising 6D pose and 3D size, is equivalent to a 3D amodal bounding box representation with free 6D pose. Given the depth point cloud at the current frame and the estimated pose from the last frame, our novel end-to-end pipeline learns to accurately update the pose. Our pipeline is composed of three modules: 1) a pose canonicalization module that normalizes the pose of the input depth point cloud; 2) RotationNet, a module that directly regresses small interframe delta rotations; and 3) CoordinateNet, a module that predicts the normalized coordinates and segmentation, enabling analytical computation of the 3D size and translation. Leveraging the small pose regime in the pose-canonicalized point clouds, our method integrates the best of both worlds by combining dense coordinate prediction and direct rotation regression, thus yielding an end-to-end differentiable pipeline optimized for 9DoF pose accuracy (without using non-differentiable RANSAC). Our extensive experiments demonstrate that our method achieves new state-of-the-art performance on category-level rigid object pose and articulated object pose benchmarks at the fastest FPS 12.

In this paper, we propose a novel two-stage context-aware network named CANet for shadow removal, in which the contextual information from non-shadow regions is transferred to shadow regions at the embedded feature spaces. At Stage-I, we propose a contextual patch matching module to generate a set of potential matching pairs of shadow and non-shadow patches. Combined with the potential contextual relationships between shadow and non-shadow regions, our well-designed contextual feature transfer (CFT) mechanism can transfer contextual information from non-shadow to shadow regions at different scales. With the reconstructed feature maps, we remove shadows at L and A/B channels separately. At Stage-II, we use an encoder-decoder to refine current results and generate the final shadow removal results. We evaluate our proposed CANet on two benchmark datasets and some real-world shadow images with complex scenes. Extensive experiment results strongly demonstrate the efficacy of our proposed CANet and exhibit superior performance to state-of-the-arts.

Temporal action localization has been one of the most popular tasks in video understanding, due to the importance of detecting action instances in videos. However, not much progress has been made on extending it to work in an online fashion, although many video related tasks can benefit by going online with the growing video streaming services. To this end, we introduce a new task called Online Temporal Action Localization (On-TAL), in which the goal is to immediately detect action instances from an untrimmed streaming video. The online setting makes the new task very challenging as the actionness decision for every frame has to be made without access to future frames and also because post-processing methods cannot be used to modify past action proposals. We propose a novel framework, Context-Aware Actionness Grouping (CAG) as a solution for On-TAL and train it with the imitation learning algorithm, which allows us to avoid sophisticated reward engineering. Evaluation of our work on THUMOS14 and Activitynet1.3 shows significant improvement over non-naive baselines, demonstrating the effectiveness of our approach. As a by-product, our method can also be used for the Online Detection of Action Start (ODAS), in which our method also outperforms previous state-of-the-art models.