ICCV 2019 论文列表

In Visual Question Answering (VQA), answers have a great correlation with question meaning and visual contents. Thus, to selectively utilize image, question and answer information, we propose a novel trilinear interaction model which simultaneously learns high level associations between these three inputs. In addition, to overcome the interaction complexity, we introduce a multimodal tensor-based PARALIND decomposition which efficiently parameterizes trilinear teraction between the three inputs. Moreover, knowledge distillation is first time applied in Free-form Opened-ended VQA. It is not only for reducing the computational cost and required memory but also for transferring knowledge from trilinear interaction model to bilinear interaction model. The extensive experiments on benchmarking datasets TDIUC, VQA-2.0, and Visual7W show that the proposed compact trilinear interaction model achieves state-of-the-art results when using a single model on all three datasets.

Few-shot learning aims to learn latent patterns from few training examples and has shown promises in practice. However, directly calculating the distances between the query image and support image in existing methods may cause ambiguity because dominant objects can locate anywhere on images. To address this issue, this paper proposes a Semantic Alignment Metric Learning (SAML) method for few-shot learning that aligns the semantically relevant dominant objects through a "collect-and-select" strategy. Specifically, we first calculate a relation matrix (RM) to "collect" the distances of each local region pairs of the 3D tensor extracted from a query image and the mean tensor of the support images. Then, the attention technique is adapted to "select" the semantically relevant pairs and put more weights on them. Afterwards, a multi-layer perceptron (MLP) is utilized to map the reweighted RMs to their corresponding similarity scores. Theoretical analysis demonstrates the generalization ability of SAML and gives a theoretical guarantee. Empirical results demonstrate that semantic alignment is achieved. Extensive experiments on benchmark datasets validate the strengths of the proposed approach and demonstrate that SAML significantly outperforms the current state-of-the-art methods. The source code is available at https://github.com/haofusheng/SAML.

The latest deep learning-based approaches have shown promising results for the challenging task of inpainting missing regions of an image. However, the existing methods often generate contents with blurry textures and distorted structures due to the discontinuity of the local pixels. From a semantic-level perspective, the local pixel discontinuity is mainly because these methods ignore the semantic relevance and feature continuity of hole regions. To handle this problem, we investigate the human behavior in repairing pictures and propose a fined deep generative model-based approach with a novel coherent semantic attention (CSA) layer, which can not only preserve contextual structure but also make more effective predictions of missing parts by modeling the semantic relevance between the holes features. The task is divided into rough, refinement as two steps and we model each step with a neural network under the U-Net architecture, where the CSA layer is embedded into the encoder of refinement step. Meanwhile, we further propose consistency loss and feature patch discriminator to stabilize the network training process and improve the details. The experiments on CelebA, Places2, and Paris StreetView datasets have validated the effectiveness of our proposed methods in image inpainting tasks and can obtain images with a higher quality as compared with the existing state-of-the-art approaches. The codes and pre-trained models will be available at https://github.com/KumapowerLIU/CSA-inpainting.

Learning how objects sound from video is challenging, since they often heavily overlap in a single audio channel. Current methods for visually-guided audio source separation sidestep the issue by training with artificially mixed video clips, but this puts unwieldy restrictions on training data collection and may even prevent learning the properties of "true" mixed sounds. We introduce a co-separation training paradigm that permits learning object-level sounds from unlabeled multi-source videos. Our novel training objective requires that the deep neural network's separated audio for similar-looking objects be consistently identifiable, while simultaneously reproducing accurate video-level audio tracks for each source training pair. Our approach disentangles sounds in realistic test videos, even in cases where an object was not observed individually during training. We obtain state-of-the-art results on visually-guided audio source separation and audio denoising for the MUSIC, AudioSet, and AV-Bench datasets.

Person re-identification (Re-ID) is an important real-world surveillance problem that entails associating a person's identity over a network of cameras. Video-based Re-ID approaches have gained significant attention recently since a video, and not just an image, is often available. In this work, we propose a novel Co-segmentation inspired video Re-ID deep architecture and formulate a Co-segmentation based Attention Module (COSAM) that activates a common set of salient features across multiple frames of a video via mutual consensus in an unsupervised manner. As opposed to most of the prior work, our approach is able to attend to person accessories along with the person. Our plug-and-play and interpretable COSAM module applied on two deep architectures (ResNet50, SE-ResNet50) outperform the state-of-the-art methods on three benchmark datasets.

Generative adversarial networks (GANs) have been successfully used for considerable computer vision tasks, especially the image-to-image translation. However, generators in these networks are of complicated architectures with large number of parameters and huge computational complexities. Existing methods are mainly designed for compressing and speeding-up deep neural networks in the classification task, and cannot be directly applied on GANs for image translation, due to their different objectives and training procedures. To this end, we develop a novel co-evolutionary approach for reducing their memory usage and FLOPs simultaneously. In practice, generators for two image domains are encoded as two populations and synergistically optimized for investigating the most important convolution filters iteratively. Fitness of each individual is calculated using the number of parameters, a discriminator-aware regularization, and the cycle consistency. Extensive experiments conducted on benchmark datasets demonstrate the effectiveness of the proposed method for obtaining compact and effective generators.

Detecting objects in aerial images is challenging for at least two reasons: (1) target objects like pedestrians are very small in pixels, making them hardly distinguished from surrounding background; and (2) targets are in general sparsely and non-uniformly distributed, making the detection very inefficient. In this paper, we address both issues inspired by observing that these targets are often clustered. In particular, we propose a Clustered Detection (ClusDet) network that unifies object clustering and detection in an end-to-end framework. The key components in ClusDet include a cluster proposal sub-network (CPNet), a scale estimation sub-network (ScaleNet), and a dedicated detection network (DetecNet). Given an input image, CPNet produces object cluster regions and ScaleNet estimates object scales for these regions. Then, each scale-normalized cluster region is fed into DetecNet for object detection. ClusDet has several advantages over previous solutions: (1) it greatly reduces the number of chips for final object detection and hence achieves high running time efficiency, (2) the cluster-based scale estimation is more accurate than previously used single-object based ones, hence effectively improves the detection for small objects, and (3) the final DetecNet is dedicated for clustered regions and implicitly models the prior context information so as to boost detection accuracy. The proposed method is tested on three popular aerial image datasets including VisDrone, UAVDT and DOTA. In all experiments, ClusDet achieves promising performance in comparison with state-of-the-art detectors.

We present a practical backend for stereo visual SLAM which can simultaneously discover individual rigid bodies and compute their motions in dynamic environments. While recent factor graph based state optimization algorithms have shown their ability to robustly solve SLAM problems by treating dynamic objects as outliers, the dynamic motions are rarely considered. In this paper, we exploit the consensus of 3D motions among the landmarks extracted from the same rigid body for clustering and estimating static and dynamic objects in a unified manner. Specifically, our algorithm builds a noise-aware motion affinity matrix upon landmarks, and uses agglomerative clustering for distinguishing those rigid bodies. Accompanied by a decoupled factor graph optimization for revising their shape and trajectory, we obtain an iterative scheme to update both cluster assignments and motion estimation reciprocally. Evaluations on both synthetic scenes and KITTI demonstrate the capability of our approach, and further experiments considering online efficiency also show the effectiveness of our method for simultaneous tracking of ego-motion and multiple objects.

Deep learning methods have shown promise in unsupervised domain adaptation, which aims to leverage a labeled source domain to learn a classifier for the unlabeled target domain with a different distribution. However, such methods typically learn a domain-invariant representation space to match the marginal distributions of the source and target domains, while ignoring their fine-level structures. In this paper, we propose Cluster Alignment with a Teacher (CAT) for unsupervised domain adaptation, which can effectively incorporate the discriminative clustering structures in both domains for better adaptation. Technically, CAT leverages an implicit ensembling teacher model to reliably discover the class-conditional structure in the feature space for the unlabeled target domain. Then CAT forces the features of both the source and the target domains to form discriminative class-conditional clusters and aligns the corresponding clusters across domains. Empirical results demonstrate that CAT achieves state-of-the-art results in several unsupervised domain adaptation scenarios.

We present ClothFlow, an appearance-flow-based generative model to synthesize clothed person for posed-guided person image generation and virtual try-on. By estimating a dense flow between source and target clothing regions, ClothFlow effectively models the geometric changes and naturally transfers the appearance to synthesize novel images as shown in Figure 1. We achieve this with a three-stage framework: 1) Conditioned on a target pose, we first estimate a person semantic layout to provide richer guidance to the generation process. 2) Built on two feature pyramid networks, a cascaded flow estimation network then accurately estimates the appearance matching between corresponding clothing regions. The resulting dense flow warps the source image to flexibly account for deformations. 3) Finally, a generative network takes the warped clothing regions as inputs and renders the target view. We conduct extensive experiments on the DeepFashion dataset for pose-guided person image generation and on the VITON dataset for the virtual try-on task. Strong qualitative and quantitative results validate the effectiveness of our method.

In this paper, we study closed-form optimal solutions to two-view triangulation with known internal calibration and pose. By formulating the triangulation problem as L-1 and L-infinity minimization of angular reprojection errors, we derive the exact closed-form solutions that guarantee global optimality under respective cost functions. To the best of our knowledge, we are the first to present such solutions. Since the angular error is rotationally invariant, our solutions can be applied for any type of central cameras, be it perspective, fisheye or omnidirectional. Our methods also require significantly less computation than the existing optimal methods. Experimental results on synthetic and real datasets validate our theoretical derivations.

Most existing text reading benchmarks make it difficult to evaluate the performance of more advanced deep learning models in large vocabularies due to the limited amount of training data. To address this issue, we introduce a new large-scale text reading benchmark dataset named Chinese Street View Text (C-SVT) with 430,000 street view images, which is at least 14 times as large as the existing Chinese text reading benchmarks. To recognize Chinese text in the wild while keeping large-scale datasets labeling cost-effective, we propose to annotate one part of the C-SVT dataset (30,000 images) in locations and text labels as full annotations and add 400,000 more images, where only the corresponding text-of-interest in the regions is given as weak annotations. To exploit the rich information from the weakly annotated data, we design a text reading network in a partially supervised learning framework, which enables to localize and recognize text, learn from fully and weakly annotated data simultaneously. To localize the best matched text proposals from weakly labeled images, we propose an online proposal matching module incorporated in the whole model, spotting the keyword regions by sharing parameters for end-to-end training. Compared with fully supervised training algorithms, this model can improve the end-to-end recognition performance remarkably by 4.03% in F-score at the same labeling cost. The proposed model can also achieve state-of-the-art results on the ICDAR 2017-RCTW dataset, which demonstrates the effectiveness of the proposed partially supervised learning framework.

In object detection, keypoint-based approaches often experience the drawback of a large number of incorrect object bounding boxes, arguably due to the lack of an additional assessment inside cropped regions. This paper presents an efficient solution that explores the visual patterns within individual cropped regions with minimal costs. We build our framework upon a representative one-stage keypoint-based detector named CornerNet. Our approach, named CenterNet, detects each object as a triplet, rather than a pair, of keypoints, which improves both precision and recall. Accordingly, we design two customized modules, cascade corner pooling, and center pooling, that enrich information collected by both the top-left and bottom-right corners and provide more recognizable information from the central regions. On the MS-COCO dataset, CenterNet achieves an AP of 47.0 %, outperforming all existing one-stage detectors by at least 4.9%. Furthermore, with a faster inference speed than the top-ranked two-stage detectors, CenterNet demonstrates a comparable performance to these detectors. Code is available at https://github.com/Duankaiwen/CenterNet.

Image-based localization (IBL) aims to estimate the 6DOF camera pose for a given query image. The camera pose can be computed from 2D-3D matches between a query image and Structure-from-Motion (SfM) models. Despite recent advances in IBL, it remains difficult to simultaneously resolve the memory consumption and match ambiguity problems of large SfM models. In this work, we propose a cascaded parallel filtering method that leverages the feature, visibility and geometry information to filter wrong matches under binary feature representation. The core idea is that we divide the challenging filtering task into two parallel tasks before deriving an auxiliary camera pose for final filtering. One task focuses on preserving potentially correct matches, while another focuses on obtaining high quality matches to facilitate subsequent more powerful filtering. Moreover, our proposed method improves the localization accuracy by introducing a quality-aware spatial reconfiguration method and a principal focal length enhanced pose estimation method. Experimental results on real-world datasets demonstrate that our method achieves very competitive localization performances in a memory-efficient manner.

In this work we propose a capsule-based approach for semi-supervised video object segmentation. Current video object segmentation methods are frame-based and often require optical flow to capture temporal consistency across frames which can be difficult to compute. To this end, we propose a video based capsule network, CapsuleVOS, which can segment several frames at once conditioned on a reference frame and segmentation mask. This conditioning is performed through a novel routing algorithm for attention-based efficient capsule selection. We address two challenging issues in video object segmentation: 1) segmentation of small objects and 2) occlusion of objects across time. The issue of segmenting small objects is addressed with a zooming module which allows the network to process small spatial regions of the video. Apart from this, the framework utilizes a novel memory module based on recurrent networks which helps in tracking objects when they move out of frame or are occluded. The network is trained end-to-end and we demonstrate its effectiveness on two benchmark video object segmentation datasets; it outperforms current offline approaches on the Youtube-VOS dataset while having a run-time that is almost twice as fast as competing methods. The code is publicly available at https://github.com/KevinDuarte/CapsuleVOS.

Learning to localize and name object instances is a fundamental problem in vision, but state-of-the-art approaches rely on expensive bounding box supervision. While weakly supervised detection (WSOD) methods relax the need for boxes to that of image-level annotations, even cheaper supervision is naturally available in the form of unstructured textual descriptions that users may freely provide when uploading image content. However, straightforward approaches to using such data for WSOD wastefully discard captions that do not exactly match object names. Instead, we show how to squeeze the most information out of these captions by training a text-only classifier that generalizes beyond dataset boundaries. Our discovery provides an opportunity for learning detection models from noisy but more abundant and freely-available caption data. We also validate our model on three classic object detection benchmarks and achieve state-of-the-art WSOD performance. Our code is available at https://github.com/yekeren/Cap2Det.

We explore the task of Canonical Surface Mapping (CSM). Specifically, given an image, we learn to map pixels on the object to their corresponding locations on an abstract 3D model of the category. But how do we learn such a mapping? A supervised approach would require extensive manual labeling which is not scalable beyond a few hand-picked categories. Our key insight is that the CSM task (pixel to 3D), when combined with 3D projection (3D to pixel), completes a cycle. Hence, we can exploit a geometric cycle consistency loss, thereby allowing us to forgo the dense manual supervision. Our approach allows us to train a CSM model for a diverse set of classes, without sparse or dense keypoint annotation, by leveraging only foreground mask labels for training. We show that our predictions also allow us to infer dense correspondence between two images, and compare the performance of our approach against several methods that predict correspondence by leveraging varying amount of supervision.

Although significant improvement has been achieved recently in 3D human pose estimation, most of the previous methods only treat a single-person case. In this work, we firstly propose a fully learning-based, camera distance-aware top-down approach for 3D multi-person pose estimation from a single RGB image. The pipeline of the proposed system consists of human detection, absolute 3D human root localization, and root-relative 3D single-person pose estimation modules. Our system achieves comparable results with the state-of-the-art 3D single-person pose estimation models without any groundtruth information and significantly outperforms previous 3D multi-person pose estimation methods on publicly available datasets. The code is available in (https://github.com/mks0601/3DMPPE_ROOTNET_RELEASE) , (https://github.com/mks0601/3DMPPE_POSENET_RELEASE).

Camera re-localization is an important but challenging task in applications like robotics and autonomous driving. Recently, retrieval-based methods have been considered as a promising direction as they can be easily generalized to novel scenes. Despite significant progress has been made, we observe that the performance bottleneck of previous methods actually lies in the retrieval module. These methods use the same features for both retrieval and relative pose regression tasks which have potential conflicts in learning. To this end, here we present a coarse-to-fine retrieval-based deep learning framework, which includes three steps, i.e., image-based coarse retrieval, pose-based fine retrieval and precise relative pose regression. With our carefully designed retrieval module, the relative pose regression task can be surprisingly simpler. We design novel retrieval losses with batch hard sampling criterion and two-stage retrieval to locate samples that adapt to the relative pose regression task. Extensive experiments show that our model (CamNet) outperforms the state-of-the-art methods by a large margin on both indoor and outdoor datasets.