CVPR 2018 Papers

Visual recognition requires rich representations that span levels from low to high, scales from small to large, and resolutions from fine to coarse. Even with the depth of features in a convolutional network, a layer in isolation is not enough: compounding and aggregating these representations improves inference of what and where. Architectural efforts are exploring many dimensions for network backbones, designing deeper or wider architectures, but how to best aggregate layers and blocks across a network deserves further attention. Although skip connections have been incorporated to combine layers, these connections have been ``shallow'' themselves, and only fuse by simple, one-step operations. We augment standard architectures with deeper aggregation to better fuse information across layers. Our deep layer aggregation structures iteratively and hierarchically merge the feature hierarchy to make networks with better accuracy and fewer parameters. Experiments across architectures and tasks show that deep layer aggregation improves recognition and resolution compared to existing branching and merging schemes.

Deep convolutional networks have become a popular tool for image generation and restoration. Generally, their excellent performance is imputed to their ability to learn realistic image priors from a large number of example images. In this paper, we show that, on the contrary, the structure of a generator network is sufficient to capture a great deal of low-level image statistics prior to any learning. In order to do so, we show that a randomly-initialized neural network can be used as a handcrafted prior with excellent results in standard inverse problems such as denoising, super-resolution, and inpainting. Furthermore, the same prior can be used to invert deep neural representations to diagnose them, and to restore images based on flash-no flash input pairs. Apart from its diverse applications, our approach highlights the inductive bias captured by standard generator network architectures. It also bridges the gap between two very popular families of image restoration methods: learning-based methods using deep convolutional networks and learning-free methods based on handcrafted image priors such as self-similarity.

This paper presents a discrepancy minimizing model to address the discrete optimization problem in hashing learning. The discrete optimization introduced by binary constraint is an NP-hard mixed integer programming problem. It is usually addressed by relaxing the binary variables into continuous variables to adapt to the gradient based learning of hashing functions, especially the training of deep neural networks. To deal with the objective discrepancy caused by relaxation, we transform the original binary optimization into differentiable optimization problem over hash functions through series expansion. This transformation decouples the binary constraint and the similarity preserving hashing function optimization. The transformed objective is optimized in a tractable alternating optimization framework with gradual discrepancy minimization. Extensive experimental results on three benchmark datasets validate the efficacy of the proposed discrepancy minimizing hashing.

Person re-identification aims at finding a person of interest in an image gallery by comparing the probe image of this person with all the gallery images. It is generally treated as a retrieval problem, where the affinities between the probe image and gallery images (P2G affinities) are used to rank the retrieved gallery images. However, most existing methods only consider P2G affinities but ignore the affinities between all the gallery images (G2G affinity). Some frameworks incorporated G2G affinities into the testing process, which is not end-to-end trainable for deep neural networks. In this paper, we propose a novel group-shuffling random walk network for fully utilizing the affinity information between gallery images in both the training and testing processes. The proposed approach aims at end-to-end refining the P2G affinities based on G2G affinity information with a simple yet effective matrix operation, which can be integrated into deep neural networks. Feature grouping and group shuffle are also proposed to apply rich supervisions for learning better person features. The proposed approach outperforms state-of-the-art methods on the Market-1501, CUHK03, and DukeMTMC datasets by large margins, which demonstrate the effectiveness of our approach.

Arguably, no single face detector fits all real-life scenarios. It is often desirable to have some built-in schemes for a face detector to automatically adapt, e.g., to a particular user's photo album (the target domain). We propose a novel face detector adaptation approach that works as long as there are representative images of the target domain no matter they are labeled or not and, more importantly, without the need of accessing the training data of the source domain. Our approach explicitly accounts for the notorious negative transfer caveat in domain adaptation thanks to a residual loss by design. Moreover, it does not incur catastrophic interference with the knowledge learned from the source domain and, therefore, the adapted face detectors maintain about the same performance as the old detectors in the original source domain. As such, our adaption approach to face detectors is analogous to the popular interpolation techniques for language models; it may opens a new direction for progressively training the face detectors domain by domain. We report extensive experimental results to verify our approach on two massively benchmarked face detectors.

This paper explores the use of extreme points in an object (left-most, right-most, top, bottom pixels) as input to obtain precise object segmentation for images and videos. We do so by adding an extra channel to the image in the input of a convolutional neural network (CNN), which contains a Gaussian centered in each of the extreme points. The CNN learns to transform this information into a segmentation of an object that matches those extreme points. We demonstrate the usefulness of this approach for guided segmentation (grabcut-style), interactive segmentation, video object segmentation, and dense segmentation annotation. We show that we obtain the most precise results to date, also with less user input, in an extensive and varied selection of benchmarks and datasets. All our models and code are publicly available on http://www.vision.ee.ethz.ch/~cvlsegmentation/dextr.

We present an end-to-end image processing framework for time-of-flight (ToF) cameras. Existing ToF image processing pipelines consist of a sequence of operations including modulated exposures, denoising, phase unwrapping and multipath interference correction. While this cascaded modular design offers several benefits, such as closed-form solutions and power-efficient processing, it also suffers from error accumulation and information loss as each module can only observe the output from its direct predecessor, resulting in erroneous depth estimates. We depart from a conventional pipeline model and propose a deep convolutional neural network architecture that recovers scene depth directly from dual-frequency, raw ToF correlation measurements. To train this network, we simulate ToF images for a variety of scenes using a time-resolved renderer, devise depth-specific losses, and apply normalization and augmentation strategies to generalize this model to real captures. We demonstrate that the proposed network can efficiently exploit the spatio-temporal structures of ToF frequency measurements, and validate the performance of the joint multipath removal, denoising and phase unwrapping method on a wide range of challenging scenes.

Spatial Transformer layers allow neural networks, at least in principle, to be invariant to large spatial transformations in image data. The model has, however, seen limited uptake as most practical implementations support only transformations that are too restricted, e.g. affine or homographic maps, and/or destructive maps, such as thin plate splines. We investigate the use of flexible diffeomorphic image transformations within such networks and demonstrate that significant performance gains can be attained over currently-used models. The learned transformations are found to be both simple and intuitive, thereby providing insights into individual problem domains. With the proposed framework, a standard convolutional neural network matches state-of-the-art results on face verification with only two extra lines of simple TensorFlow code.

The goal of our work is to complete the depth channel of an RGB-D image. Commodity-grade depth cameras often fail to sense depth for shiny, bright, transparent, and distant surfaces. To address this problem, we train a deep network that takes an RGB image as input and predicts dense surface normals and occlusion boundaries. Those predictions are then combined with raw depth observations provided by the RGB-D camera to solve for depths for all pixels, including those missing in the original observation. This method was chosen over others (e.g., inpainting depths directly) as the result of extensive experiments with a new depth completion benchmark dataset, where holes are filled in training data through the rendering of surface reconstructions created from multiview RGB-D scans. Experiments with different network inputs, depth representations, loss functions, optimization methods, inpainting methods, and deep depth estimation networks show that our proposed approach provides better depth completions than these alternatives.

In this paper, we consider the problem of grouping a collection of unconstrained face images in which the number of subjects is not known. We propose an unsupervised clustering algorithm called Deep Density Clustering (DDC) which is based on measuring density affinities between local neighborhoods in the feature space. By learning the minimal covering sphere for each neighborhood, information about the underlying structure is encapsulated. The encapsulation is also capable of locating high-density region of the neighborhood, which aids in measuring the neighborhood similarity. We theoretically show that the encapsulation asymptotically converges to a Parzen window density estimator. Our experiments show that DDC is a superior candidate for clustering unconstrained faces when the number of subjects is unknown. Unlike conventional linkage and density-based methods that are sensitive to the selection operating points, DDC attains more consistent and improved performance. Furthermore, the density-aware property reduces the difficulty in finding appropriate operating points.

Cross-media retrieval is a research hotspot in multimedia area, which aims to perform retrieval across different media types such as image and text. The performance of existing methods usually relies on labeled data for model training. However, cross-media data is very labor consuming to collect and label, so how to transfer valuable knowledge in existing data to new data is a key problem towards application. For achieving the goal, this paper proposes deep cross-media knowledge transfer (DCKT) approach, which transfers knowledge from a large-scale cross-media dataset to promote the model training on another small-scale cross-media dataset. The main contributions of DCKT are: (1) Two-level transfer architecture is proposed to jointly minimize the media-level and correlation-level domain discrepancies, which allows two important and complementary aspects of knowledge to be transferred: intra-media semantic and inter-media correlation knowledge. It can enrich the training information and boost the retrieval accuracy. (2) Progressive transfer mechanism is proposed to iteratively select training samples with ascending transfer difficulties, via the metric of cross-media domain consistency with adaptive feedback. It can drive the transfer process to gradually reduce vast cross-media domain discrepancy, so as to enhance the robustness of model training. For verifying the effectiveness of DCKT, we take the large-scale dataset XMediaNet as source domain, and 3 widely-used datasets as target domain for cross-media retrieval. Experimental results show that DCKT achieves promising improvement on retrieval accuracy.

Facial age estimation from a face image is an important yet very challenging task in computer vision, since humans with different races and/or genders, exhibit quite different patterns in their facial aging processes. To deal with the influence of race and gender, previous methods perform age estimation within each population separately. In practice, however, it is often very difficult to collect and label sufficient data for each population. Therefore, it would be helpful to exploit an existing large labeled dataset of one (source) population to improve the age estimation performance on another (target) population with only a small labeled dataset available. In this work, we propose a Deep Cross-Population (DCP) age estimation model to achieve this goal. In particular, our DCP model develops a two-stage training strategy. First, a novel cost-sensitive multi-task loss function is designed to learn transferable aging features by training on the source population. Second, a novel order-preserving pair-wise loss function is designed to align the aging features of the two populations. By doing so, our DCP model can transfer the knowledge encoded in the source population to the target population. Extensive experiments on the two of the largest benchmark datasets show that our DCP model outperforms several strong baseline methods and many state-of-the-art methods.

Most existing unsupervised domain adaptation (UDA) methods are based upon the assumption that source labeled data come from an identical underlying distribution. Whereas in practical scenario, labeled instances are typically collected from diverse sources. Moreover, those sources may not completely share their categories, which further brings a category shift challenge to multi-source (unsupervised) domain adaptation (MDA). In this paper, we propose a deep cocktail network (DCTN), to battle the domain and category shifts among multiple sources. Motivated by the theoretical results in cite{mansour2009domain}, the target distribution can be represented as the weighted combination of source distributions, and, the training of MDA via DCTN is then performed as two alternating steps: i) It deploys multi-way adversarial learning to minimize the discrepancy between the target and each of the multiple source domains, which also obtains the source-specific perplexity scores to denote the possibilities that a target sample belongs to different source domains. ii) The multi-source category classifiers are integrated with the perplexity scores to classify target sample, and the pseudo-labeled target samples together with source samples are utilized to update the multi-source category classifier and the representation module. We evaluate DCTN in three domain adaptation benchmarks, which clearly demonstrate the superiority of our framework.

Due to its computation efficiency and retrieval quality, hashing has been widely applied to approximate nearest neighbor search for large-scale image retrieval, while deep hashing further improves the retrieval quality by end-to-end representation learning and hash coding. With compact hash codes, Hamming space retrieval enables the most efficient constant-time search that returns data points within a given Hamming radius to each query, by hash table lookups instead of linear scan. However, subject to the weak capability of concentrating relevant images to be within a small Hamming ball due to mis-specified loss functions, existing deep hashing methods may underperform for Hamming space retrieval. This work presents Deep Cauchy Hashing (DCH), a novel deep hashing model that generates compact and concentrated binary hash codes to enable efficient and effective Hamming space retrieval. The main idea is to design a pairwise cross-entropy loss based on Cauchy distribution, which penalizes significantly on similar image pairs with Hamming distance larger than the given Hamming radius threshold. Comprehensive experiments demonstrate that DCH can generate highly concentrated hash codes and yield state-of-the-art Hamming space retrieval performance on three datasets, NUS-WIDE, CIFAR-10, and MS-COCO.

The feed-forward architectures of recently proposed deep super-resolution networks learn representations of low-resolution inputs, and the non-linear mapping from those to high-resolution output. However, this approach does not fully address the mutual dependencies of low- and high-resolution images. We propose Deep Back-Projection Networks (DBPN), that exploit iterative up- and down-sampling layers, providing an error feedback mechanism for projection errors at each stage. We construct mutually-connected up- and down-sampling stages each of which represents different types of image degradation and high-resolution components. We show that extending this idea to allow concatenation of features across up- and down-sampling stages (Dense DBPN) allows us to reconstruct further improve super-resolution, yielding superior results and in particular establishing new state of the art results for large scaling factors such as 8x across multiple data sets.

Most existing subspace clustering methods hinge on self-expression of handcrafted representations and are unaware of potential clustering errors. Thus they perform unsatisfactorily on real data with complex underlying subspaces. To solve this issue, we propose a novel deep adversarial subspace clustering (DASC) model, which learns more favorable sample representations by deep learning for subspace clustering, and more importantly introduces adversarial learning to supervise sample representation learning and subspace clustering. Specifically, DASC consists of a subspace clustering generator and a quality-verifying discriminator, which learn against each other. The generator produces subspace estimation and sample clustering. The discriminator evaluates current clustering performance by inspecting whether the re-sampled data from estimated subspaces have consistent subspace properties, and supervises the generator to progressively improve subspace clustering. Experimental results on the handwritten recognition, face and object clustering tasks demonstrate the advantages of DASC over shallow and few deep subspace clustering models. Moreover, to our best knowledge, this is the first successful application of GAN-alike model for unsupervised subspace clustering, which also paves the way for deep learning to solve other unsupervised learning problems.

Learning an effective distance metric between image pairs plays an important role in visual analysis, where the training procedure largely relies on hard negative samples. However, hard negatives in the training set usually account for the tiny minority, which may fail to fully describe the distribution of negative samples close to the margin. In this paper, we propose a deep adversarial metric learning (DAML) framework to generate synthetic hard negatives from the observed negative samples, which is widely applicable to supervised deep metric learning methods. Different from existing metric learning approaches which simply ignore numerous easy negatives, the proposed DAML exploits them to generate potential hard negatives adversary to the learned metric as complements. We simultaneously train the hard negative generator and feature embedding in an adversarial manner, so that more precise distance metrics can be learned with adequate and targeted synthetic hard negatives. Extensive experimental results on three benchmark datasets including CUB-200-2011, Cars196 and Stanford Online Products show that DAML effectively boosts the performance of existing deep metric learning approaches through adversarial learning.

Inner product-based convolution has been a central component of convolutional neural networks (CNNs) and the key to learning visual representations. Inspired by the observation that CNN-learned features are naturally decoupled with the norm of features corresponding to the intra-class variation and the angle corresponding to the semantic difference, we propose a generic decoupled learning framework which models the intra-class variation and semantic difference independently. Specifically, we first reparametrize the inner product to a decoupled form and then generalize it to the decoupled convolution operator which serves as the building block of our decoupled networks. We present several effective instances of the decoupled convolution operator. Each decoupled operator is well motivated and has an intuitive geometric interpretation. Based on these decoupled operators, we further propose to directly learn the operator from data. Extensive experiments show that such decoupled reparameterization renders significant performance gain with easier convergence and stronger robustness.

Batch Normalization (BN) is capable of accelerating the training of deep models by centering and scaling activations within mini-batches. In this work, we propose Decorrelated Batch Normalization (DBN), which not just centers and scales activations but whitens them. We explore multiple whitening techniques, and find that PCA whitening causes a problem we call stochastic axis swapping, which is detrimental to learning. We show that ZCA whitening does not suffer from this problem, permitting successful learning. DBN retains the desirable qualities of BN and further improves BN's optimization efficiency and generalization ability. We design comprehensive experiments to show that DBN can improve the performance of BN on multilayer perceptrons and convolutional neural networks. Furthermore, we consistently improve the accuracy of residual networks on CIFAR-10, CIFAR-100, and ImageNet.