ICCV 2019 论文列表

Optical flow in heavy rainy scenes is challenging due to the presence of both rain steaks and rain veiling effect, which break the existing optical flow constraints. Concerning this, we propose a deep-learning based optical flow method designed to handle heavy rain. We introduce a feature multiplier in our network that transforms the features of an image affected by the rain veiling effect into features that are less affected by it, which we call veiling-invariant features. We establish a new mapping operation in the feature space to produce streak-invariant features. The operation is based on a feature pyramid structure of the input images, and the basic idea is to preserve the chromatic features of the background scenes while canceling the rain-streak patterns. Both the veiling-invariant and streak-invariant features are computed and optimized automatically based on the the accuracy of our optical flow estimation. Our network is end-to-end, and handles both rain streaks and the veiling effect in an integrated framework. Extensive experiments show the effectiveness of our method, which outperforms the state of the art method and other baseline methods. We also show that our network can robustly maintain good performance on clean (no rain) images even though it is trained under rain image data.

Racial bias is an important issue in biometric, but has not been thoroughly studied in deep face recognition. In this paper, we first contribute a dedicated dataset called Racial Faces in-the-Wild (RFW) database, on which we firmly validated the racial bias of four commercial APIs and four state-of-the-art (SOTA) algorithms. Then, we further present the solution using deep unsupervised domain adaptation and propose a deep information maximization adaptation network (IMAN) to alleviate this bias by using Caucasian as source domain and other races as target domains. This unsupervised method simultaneously aligns global distribution to decrease race gap at domain-level, and learns the discriminative target representations at cluster level. A novel mutual information loss is proposed to further enhance the discriminative ability of network output without label information. Extensive experiments on RFW, GBU, and IJB-A databases show that IMAN successfully learns features that generalize well across different races and across different databases.

In this work, we introduce the task of 3D object instance re-localization (RIO): given one or multiple objects in an RGB-D scan, we want to estimate their corresponding 6DoF poses in another 3D scan of the same environment taken at a later point in time. We consider RIO a particularly important task in 3D vision since it enables a wide range of practical applications, including AI-assistants or robots that are asked to find a specific object in a 3D scene. To address this problem, we first introduce 3RScan, a novel dataset and benchmark, which features 1482 RGB-D scans of 478 environments across multiple time steps. Each scene includes several objects whose positions change over time, together with ground truth annotations of object instances and their respective 6DoF mappings among re-scans. Automatically finding 6DoF object poses leads to a particular challenging feature matching task due to varying partial observations and changes in the surrounding context. To this end, we introduce a new data-driven approach that efficiently finds matching features using a fully-convolutional 3D correspondence network operating on multiple spatial scales. Combined with a 6DoF pose optimization, our method outperforms state-of-the-art baselines on our newly-established benchmark, achieving an accuracy of 30.58%.

RGB-Infrared (IR) person re-identification is an important and challenging task due to large cross-modality variations between RGB and IR images. Most conventional approaches aim to bridge the cross-modality gap with feature alignment by feature representation learning. Different from existing methods, in this paper, we propose a novel and end-to-end Alignment Generative Adversarial Network (AlignGAN) for the RGB-IR RE-ID task. The proposed model enjoys several merits. First, it can exploit pixel alignment and feature alignment jointly. To the best of our knowledge, this is the first work to model the two alignment strategies jointly for the RGB-IR RE-ID problem. Second, the proposed model consists of a pixel generator, a feature generator and a joint discriminator. By playing a min-max game among the three components, our model is able to not only alleviate the cross-modality and intra-modality variations, but also learn identity-consistent features. Extensive experimental results on two standard benchmarks demonstrate that the proposed model performs favourably against state-of-the-art methods. Especially, on SYSU-MM01 dataset, our model can achieve an absolute gain of 15.4% and 12.9% in terms of Rank-1 and mAP.

Despite online learning (OL) techniques have boosted the performance of semi-supervised video object segmentation (VOS) methods, the huge time costs of OL greatly restricts their practicality. Matching based and propagation based methods run at a faster speed by avoiding OL techniques. However, they are limited by sub-optimal accuracy, due to mismatching and drifting problems. In this paper, we develop a real-time yet very accurate Ranking Attention Network (RANet) for VOS. Specifically, to integrate the insights of matching based and propagation based methods, we employ an encoder-decoder framework to learn pixel-level similarity and segmentation in an end-to-end manner. To better utilize the similarity maps, we propose a novel ranking attention module, which automatically ranks and selects these maps for fine-grained VOS performance. Experiments on DAVIS16 and DAVIS17 datasets show that our RANet achieves the best speed-accuracy trade-off, e.g., with 33 milliseconds per frame and J&F=85.5% on DAVIS16. With OL, our RANet reaches J&F=87.1% on DAVIS16, exceeding state-of-the-art VOS methods. The code can be found at https://github.com/Storife/RANet.

The image lines projected from parallel 3D lines intersect at a common point called the vanishing point (VP). Manhattan world holds for the scenes with three orthogonal VPs. In Manhattan world, given several lines in a calibrated image, we aim at clustering them by three unknown-but-sought VPs. The VP estimation can be reformulated as computing the rotation between the Manhattan frame and the camera frame. To compute this rotation, state-of-the-art methods are based on either data sampling or parameter search, and they fail to guarantee the accuracy and efficiency simultaneously. In contrast, we propose to hybridize these two strategies. We first compute two degrees of freedom (DOF) of the above rotation by two sampled image lines, and then search for the optimal third DOF based on the branch-and-bound. Our sampling accelerates our search by reducing the search space and simplifying the bound computation. Our search is not sensitive to noise and achieves quasi-global optimality in terms of maximizing the number of inliers. Experiments on synthetic and real-world images showed that our method outperforms state-of-the-art approaches in terms of accuracy and/or efficiency.

We present a new quasi-affine reconstruction of a scene and its application to camera self-calibration. We refer to this reconstruction as QUARCH (QUasi-Affine Reconstruction with respect to Camera centers and the Hodographs of horopters). A QUARCH can be obtained by solving a semidefinite programming problem when, (i) the images have been captured by a moving camera with constant intrinsic parameters, and (ii) a vague knowledge of the relative orientation (under or over 120 degrees) between camera pairs is available. The resulting reconstruction comes close enough to an affine one allowing thus an easy upgrade of the QUARCH to its affine and metric counterparts. We also present a constrained Levenberg-Marquardt method for nonlinear optimization subject to Linear Matrix Inequality (LMI) constraints so as to ensure that the QUARCH LMIs are satisfied during optimization. Experiments with synthetic and real data show the benefits of QUARCH in reliably obtaining a metric reconstruction.

One-shot image segmentation aims to undertake the segmentation task of a novel class with only one training image available. The difficulty lies in that image segmentation has structured data representations, which yields a many-to-many message passing problem. Previous methods often simplify it to a one-to-many problem by squeezing support data to a global descriptor. However, a mixed global representation drops the data structure and information of individual elements. In this paper, we propose to model structured segmentation data with graphs and apply attentive graph reasoning to propagate label information from support data to query data. The graph attention mechanism could establish the element-to-element correspondence across structured data by learning attention weights between connected graph nodes. To capture correspondence at different semantic levels, we further propose a pyramid-like structure that models different sizes of image regions as graph nodes and undertakes graph reasoning at different levels. Experiments on PASCAL VOC 2012 dataset demonstrate that our proposed network significantly outperforms the baseline method and leads to new state-of-the-art performance on 1-shot and 5-shot segmentation benchmarks.

In this work, we propose a novel crowd counting network that progressively generates crowd density maps via residual error estimation. The proposed method uses VGG16 as the backbone network and employs density map generated by the final layer as a coarse prediction to refine and generate finer density maps in a progressive fashion using residual learning. Additionally, the residual learning is guided by an uncertainty-based confidence weighting mechanism that permits the flow of only high-confidence residuals in the refinement path. The proposed Confidence Guided Deep Residual Counting Network (CG-DRCN) is evaluated on recent complex datasets, and it achieves significant improvements in errors. Furthermore, we introduce a new large scale unconstrained crowd counting dataset (JHU-CROWD) that is 2.8 larger than the most recent crowd counting datasets in terms of the number of images. It contains 4,250 images with 1.11 million annotations. In comparison to existing datasets, the proposed dataset is collected under a variety of diverse scenarios and environmental conditions. Specifically, the dataset includes several images with weather-based degradations and illumination variations in addition to many distractor images, making it a very challenging dataset. Additionally, the dataset consists of rich annotations at both image-level and head-level. Several recent methods are evaluated and compared on this dataset.

In this work we propose a model that can manipulate individual visual attributes of objects in a real scene using examples of how respective attribute manipulations affect the output of a simulation. As an example, we train our model to manipulate the expression of a human face using nonphotorealistic 3D renders of a face with varied expression. Our model manages to preserve all other visual attributes of a real face, such as head orientation, even though this and other attributes are not labeled in either real or synthetic domain. Since our model learns to manipulate a specific property in isolation using only "synthetic demonstrations" of such manipulations without explicitly provided labels, it can be applied to shape, texture, lighting, and other properties that are difficult to measure or represent as real-valued vectors. We measure the degree to which our model preserves other attributes of a real image when a single specific attribute is manipulated. We use digit datasets to analyze how discrepancy in attribute distributions affects the performance of our model, and demonstrate results in a far more difficult setting: learning to manipulate real human faces using nonphotorealistic 3D renders.

Compressing large Neural Networks (NN) by quantizing the parameters, while maintaining the performance is highly desirable due to reduced memory and time complexity. In this work, we cast NN quantization as a discrete labelling problem, and by examining relaxations, we design an efficient iterative optimization procedure that involves stochastic gradient descent followed by a projection. We prove that our simple projected gradient descent approach is, in fact, equivalent to a proximal version of the well-known mean-field method. These findings would allow the decades-old and theoretically grounded research on MRF optimization to be used to design better network quantization schemes. Our experiments on standard classification datasets (MNIST, CIFAR10/100, TinyImageNet) with convolutional and residual architectures show that our algorithm obtains fully-quantized networks with accuracies very close to the floating-point reference networks.

The Progressive-X algorithm, Prog-X in short, is proposed for geometric multi-model fitting. The method interleaves sampling and consolidation of the current data interpretation via repetitive hypothesis proposal, fast rejection, and integration of the new hypothesis into the kept instance set by labeling energy minimization. Due to exploring the data progressively, the method has several beneficial properties compared with the state-of-the-art. First, a clear criterion, adopted from RANSAC, controls the termination and stops the algorithm when the probability of finding a new model with a reasonable number of inliers falls below a threshold. Second, Prog-X is an any-time algorithm. Thus, whenever is interrupted, e.g. due to a time limit, the returned instances cover real and, likely, the most dominant ones. The method is superior to the state-of-the-art in terms of accuracy in both synthetic experiments and on publicly available real-world datasets for homography, two-view motion, and motion segmentation.

Transferring image-based object detectors to the domain of videos remains a challenging problem. Previous efforts mostly exploit optical flow to propagate features across frames, aiming to achieve a good trade-off between accuracy and efficiency. However, introducing an extra model to estimate optical flow can significantly increase the overall model size. The gap between optical flow and high-level features can also hinder it from establishing spatial correspondence accurately. Instead of relying on optical flow, this paper proposes a novel module called Progressive Sparse Local Attention (PSLA), which establishes the spatial correspondence between features across frames in a local region with progressively sparser stride and uses the correspondence to propagate features. Based on PSLA, Recursive Feature Updating (RFU) and Dense Feature Transforming (DenseFT) are proposed to model temporal appearance and enrich feature representation respectively in a novel video object detection framework. Experiments on ImageNet VID show that our method achieves the best accuracy compared to existing methods with smaller model size and acceptable runtime speed.

Inpainting methods aim to restore missing parts of corrupted images and play a critical role in many computer vision applications, such as object removal and image restoration. Although existing methods perform well on images with small holes, restoring large holes remains elusive. To address this issue, this paper proposes a Progressive Reconstruction of Visual Structure (PRVS) network that progressively reconstructs the structures and the associated visual feature. Specifically, we design a novel Visual Structure Reconstruction (VSR) layer to entangle reconstructions of the visual structure and visual feature, which benefits each other by sharing parameters. We repeatedly stack four VSR layers in both encoding and decoding stages of a U-Net like architecture to form the generator of a generative adversarial network (GAN) for restoring images with either small or large holes. We prove the generalization error upper bound of the PRVS network is O(1sqrt(N)), which theoretically guarantees its performance. Extensive empirical evaluations and comparisons on Places2, Paris Street View and CelebA datasets validate the strengths of the proposed approach and demonstrate that the model outperforms current state-of-the-art methods. The source code package is available at https://github.com/jingyuanli001/PRVS-Image-Inpainting.

Most previous fusion strategies either fail to fully utilize temporal information or cost too much time, and how to effectively fuse temporal information from consecutive frames plays an important role in video super-resolution (SR). In this study, we propose a novel progressive fusion network for video SR, which is designed to make better use of spatio-temporal information and is proved to be more efficient and effective than the existing direct fusion, slow fusion or 3D convolution strategies. Under this progressive fusion framework, we further introduce an improved non-local operation to avoid the complex motion estimation and motion compensation (ME&MC) procedures as in previous video SR approaches. Extensive experiments on public datasets demonstrate that our method surpasses state-of-the-art with 0.96 dB in average, and runs about 3 times faster, while requires only about half of the parameters.

Recently, differentiable search methods have made major progress in reducing the computational costs of neural architecture search. However, these approaches often report lower accuracy in evaluating the searched architecture or transferring it to another dataset. This is arguably due to the large gap between the architecture depths in search and evaluation scenarios. In this paper, we present an efficient algorithm which allows the depth of searched architectures to grow gradually during the training procedure. This brings two issues, namely, heavier computational overheads and weaker search stability, which we solve using search space approximation and regularization, respectively. With a significantly reduced search time ( 7 hours on a single GPU), our approach achieves state-of-the-art performance on both the proxy dataset (CIFAR10 or CIFAR100) and the target dataset (ImageNet). Code is available at https://github.com/chenxin061/pdarts

Humans are capable of building holistic representations for images at various levels, from local objects, to pairwise relations, to global structures. The interpretation of structures involves reasoning over repetition and symmetry of the objects in the image. In this paper, we present the Program-Guided Image Manipulator (PG-IM), inducing neuro-symbolic program-like representations to represent and manipulate images. Given an image, PG-IM detects repeated patterns, induces symbolic programs, and manipulates the image using a neural network that is guided by the program. PG-IM learns from a single image, exploiting its internal statistics. Despite trained only on image inpainting, PG-IM is directly capable of extrapolation and regularity editing in a unified framework. Extensive experiments show that PG-IM achieves superior performance on all the tasks.

Embedding methods have achieved success in face recognition by comparing facial features in a latent semantic space. However, in a fully unconstrained face setting, the facial features learned by the embedding model could be ambiguous or may not even be present in the input face, leading to noisy representations. We propose Probabilistic Face Embeddings (PFEs), which represent each face image as a Gaussian distribution in the latent space. The mean of the distribution estimates the most likely feature values while the variance shows the uncertainty in the feature values. Probabilistic solutions can then be naturally derived for matching and fusing PFEs using the uncertainty information. Empirical evaluation on different baseline models, training datasets and benchmarks show that the proposed method can improve the face recognition performance of deterministic embeddings by converting them into PFEs. The uncertainties estimated by PFEs also serve as good indicators of the potential matching accuracy, which are important for a risk-controlled recognition system.

With excellent representation power for complex data, deep neural networks (DNNs) based approaches are state-of-the-art for ordinal regression problem which aims to classify instances into ordinal categories. However, DNNs are not able to capture uncertainties and produce probabilistic interpretations. As a probabilistic model, Gaussian Processes (GPs) on the other hand offers uncertainty information, which is nonetheless lack of scalability for large datasets. This paper adapts traditional GPs regression for ordinal regression problem by using both conjugate and non-conjugate ordinal likelihood. Based on that, it proposes a deep neural network with a GPs layer on the top, which is trained end-to-end by the stochastic gradient descent method for both neural network parameters and GPs parameters. The parameters in the ordinal likelihood function are learned as neural network parameters so that the proposed framework is able to produce fitted likelihood functions for training sets and make probabilistic predictions for test points. Experimental results on three real-world benchmarks -- image aesthetics rating, historical image grading and age group estimation -- demonstrate that in terms of mean absolute error, the proposed approach outperforms state-of-the-art ordinal regression approaches and provides the confidence for predictions.