ICCV 2015 论文列表

We present an approach to utilize large amounts of web data for learning CNNs. Specifically inspired by curriculum learning, we present a two-step approach for CNN training. First, we use easy images to train an initial visual representation. We then use this initial CNN and adapt it to harder, more realistic images by leveraging the structure of data and categories. We demonstrate that our two-stage CNN outperforms a fine-tuned CNN trained on ImageNet on Pascal VOC 2012. We also demonstrate the strength of webly supervised learning by localizing objects in web images and training a R-CNN style detector. It achieves the best performance on VOC 2007 where no VOC training data is used. Finally, we show our approach is quite robust to noise and performs comparably even when we use image search results from March 2013 (pre-CNN image search era).

Large scale duplicate detection, clustering and mining of documents or images has been conventionally treated with seed detection via hashing, followed by seed growing heuristics using fast search. Principled clustering methods, especially kernelized and spectral ones, have higher complexity and are difficult to scale above millions. Under the assumption of documents or images embedded in Euclidean space, we revisit recent advances in approximate k-means variants, and borrow their best ingredients to introduce a new one, inverted-quantized k-means (IQ-means). Key underlying concepts are quantization of data points and multi-index based inverted search from centroids to cells. Its quantization is a form of hashing and analogous to seed detection, while its updates are analogous to seed growing, yet principled in the sense of distortion minimization. We further design a dynamic variant that is able to determine the number of clusters k in a single run at nearly zero additional cost. Combined with powerful deep learned representations, we achieve clustering of a 100 million image collection on a single machine in less than one hour.

We present a weakly-supervised approach to semantic segmentation. The goal is to assign pixel-level labels given only partial information, for example, image-level labels. This is an important problem in many application scenarios where it is difficult to get accurate segmentation or not feasible to obtain detailed annotations. The proposed approach starts with an initial coarse segmentation, followed by a spectral clustering approach that groups related image parts into communities. A community-driven graph is then constructed that captures spatial and feature relationships between communities while a label graph captures correlations between image labels. Finally, mapping the image level labels to appropriate communities is formulated as a convex optimization problem. The proposed approach does not require location information for image level labels and can be trained using partially labeled datasets. Compared to the state-of-the-art weakly supervised approaches, we achieve a significant performance improvement of 9% on MSRC-21 dataset and 11% on LabelMe dataset, while being more than 300 times faster.

This work explores the use of spatial context as a source of free and plentiful supervisory signal for training a rich visual representation. Given only a large, unlabeled image collection, we extract random pairs of patches from each image and train a convolutional neural net to predict the position of the second patch relative to the first. We argue that doing well on this task requires the model to learn to recognize objects and their parts. We demonstrate that the feature representation learned using this within-image context indeed captures visual similarity across images. For example, this representation allows us to perform unsupervised visual discovery of objects like cats, people, and even birds from the Pascal VOC 2011 detection dataset. Furthermore, we show that the learned ConvNet can be used in the R-CNN framework and provides a significant boost over a randomly-initialized ConvNet, resulting in state-of-the-art performance among algorithms which use only Pascal-provided training set annotations.

We propose an entirely data-driven approach to estimating the 3D pose of a hand given a depth image. We show that we can correct the mistakes made by a Convolutional Neural Network trained to predict an estimate of the 3D pose by using a feedback loop. The components of this feedback loop are also Deep Networks, optimized using training data. They remove the need for fitting a 3D model to the input data, which requires both a carefully designed fitting function and algorithm. We show that our approach outperforms state-of-the-art methods, and is efficient as our implementation runs at over 400 fps on a single GPU.

This paper presents a novel 3D modeling framework that reconstructs an indoor scene as a structured model from panorama RGBD images. A scene geometry is represented as a graph, where nodes correspond to structural elements such as rooms, walls, and objects. The approach devises a structure grammar that defines how a scene graph can be manipulated. The grammar then drives a principled new reconstruction algorithm, where the grammar rules are sequentially applied to recover a structured model. The paper also proposes a new room segmentation algorithm and an offset-map reconstruction algorithm that are used in the framework and can enforce architectural shape priors far beyond existing state-of-the-art. The structured scene representation enables a variety of novel applications, ranging from indoor scene visualization, automated floorplan generation, Inverse-CAD, and more. We have tested our framework and algorithms on six synthetic and five real datasets with qualitative and quantitative evaluations.

This paper addresses semantic image segmentation by incorporating rich information into Markov Random Field (MRF), including high-order relations and mixture of label contexts. Unlike previous works that optimized MRFs using iterative algorithm, we solve MRF by proposing a Convolutional Neural Network (CNN), namely Deep Parsing Network (DPN), which enables deterministic end-to-end computation in a single forward pass. Specifically, DPN extends a contemporary CNN architecture to model unary terms and additional layers are carefully devised to approximate the mean field algorithm (MF) for pairwise terms. It has several appealing properties. First, different from the recent works that combined CNN and MRF, where many iterations of MF were required for each training image during back-propagation, DPN is able to achieve high performance by approximating one iteration of MF. Second, DPN represents various types of pairwise terms, making many existing works as its special cases. Third, DPN makes MF easier to be parallelized and speeded up in Graphical Processing Unit (GPU). DPN is thoroughly evaluated on the PASCAL VOC 2012 dataset, where a single DPN model yields a new state-of-the-art segmentation accuracy of 77.5%.

Unsupervised and weakly-supervised visual learning in large image collections are critical in order to avoid the time-consuming and error-prone process of manual labeling. Standard approaches rely on methods like multiple-instance learning or graphical models, which can be computationally intensive and sensitive to initialization. On the other hand, simpler component analysis or clustering methods usually cannot achieve meaningful invariances or semantic interpretability. To address the issues of previous work, we present a simple but effective method called Semantic Component Analysis (SCA), which provides a decomposition of images into semantic components. Unsupervised SCA decomposes additive image representations into spatially-meaningful visual components that naturally correspond to object categories. Using an overcomplete representation that allows for rich instance-level constraints and spatial priors, SCA gives improved results and more interpretable components in comparison to traditional matrix factorization techniques. If weakly-supervised information is available in the form of image-level tags, SCA factorizes a set of images into semantic groups of superpixels. We also provide qualitative connections to traditional methods for component analysis (e.g. Grassmann averages, PCA, and NMF). The effectiveness of our approach is validated through synthetic data and on the MSRC2 and Sift Flow datasets, demonstrating competitive results in unsupervised and weakly-supervised semantic segmentation.

We introduce Segment-Phrase Table (SPT), a large collection of bijective associations between textual phrases and their corresponding segmentations. Leveraging recent progress in object recognition and natural language semantics, we show how we can successfully build a high-quality segment-phrase table using minimal human supervision. More importantly, we demonstrate the unique value unleashed by this rich bimodal resource, for both vision as well as natural language understanding. First, we show that fine-grained textual labels facilitate contextual reasoning that helps in satisfying semantic constraints across image segments. This feature enables us to achieve state-of-the-art segmentation results on benchmark datasets. Next, we show that the association of high-quality segmentations to textual phrases aids in richer semantic understanding and reasoning of these textual phrases. Leveraging this feature, we motivate the problem of visual entailment and visual paraphrasing, and demonstrate its utility on a large dataset.

Markov random fields are widely used to model many computer vision problems that can be cast in an energy minimization framework composed of unary and pairwise potentials. While computationally tractable discrete optimizers such as Graph Cuts and belief propagation (BP) exist for multi-label discrete problems, they still face prohibitively high computational challenges when the labels reside in a huge or very densely sampled space. Integrating key ideas from PatchMatch of effective particle propagation and resampling, PatchMatch belief propagation (PMBP) has been demonstrated to have good performance in addressing continuous labeling problems and runs orders of magnitude faster than Particle BP (PBP). However, the quality of the PMBP solution is tightly coupled with the local window size, over which the raw data cost is aggregated to mitigate ambiguity in the data constraint. This dependency heavily influences the overall complexity, increasing linearly with the window size. This paper proposes a novel algorithm called sped-up PMBP (SPM-BP) to tackle this critical computational bottleneck and speeds up PMBP by 50-100 times. The crux of SPM-BP is on unifying efficient filter-based cost aggregation and message passing with PatchMatch-based particle generation in a highly effective way. Though simple in its formulation, SPM-BP achieves superior performance for sub-pixel accurate stereo and optical-flow on benchmark datasets when compared with more complex and task-specific approaches.

This paper deals with the problem of registering a known structured 3D scene and its metric Structure-from-Motion (SfM) counterpart. The proposed work relies on a prior plane segmentation of the 3D scene and aligns the data obtained from both modalities by solving the point-to-plane assignment problem. An inliers-maximization approach within a Branch-and-Bound (BnB) search scheme is adopted. For the first time in this paper, a Sum-of-Squares optimization theory framework is employed for identifying point-to-plane mismatches (i.e. outliers) with certainty. This allows us to iteratively build potential inliers sets and converge to the solution satisfied by the largest number of point-to-plane assignments. Furthermore, our approach is boosted by new plane visibility conditions which are also introduced in this paper. Using this framework, we solve the registration problem in two cases: (i) a set of putative point-to-plane correspondences (with possibly overwhelmingly many outliers) is given as input and (ii) no initial correspondences are given. In both cases, our approach yields outstanding results in terms of robustness and optimality.

We present an approach to nonrigid registration of 3D surfaces. We cast isometric embedding as MRF optimization and apply efficient global optimization algorithms based on linear programming relaxations. The Markov random field perspective suggests a natural connection with robust statistics and motivates robust forms of the intrinsic distortion functional. Our approach outperforms a large body of prior work by a significant margin, increasing registration precision on real data by a factor of 3.

Object viewpoint estimation from 2D images is an essential task in computer vision. However, two issues hinder its progress: scarcity of training data with viewpoint annotations, and a lack of powerful features. Inspired by the growing availability of 3D models, we propose a framework to address both issues by combining render-based image synthesis and CNNs (Convolutional Neural Networks). We believe that 3D models have the potential in generating a large number of images of high variation, which can be well exploited by deep CNN with a high learning capacity. Towards this goal, we propose a scalable and overfit-resistant image synthesis pipeline, together with a novel CNN specifically tailored for the viewpoint estimation task. Experimentally, we show that the viewpoint estimation from our pipeline can significantly outperform state-of-the-art methods on PASCAL 3D+ benchmark.

Visual distortions on images caused by bad weather conditions can have a negative impact on the performance of many outdoor vision systems. One often seen bad weather is rain which causes significant yet complex local intensity fluctuations in images. The paper aims at developing an effective algorithm to remove visual effects of rain from a single rainy image, i.e. separate the rain layer and the de-rained image layer from an rainy image. Built upon a non-linear generative model of rainy image, namely screen blend mode, we proposed a dictionary learning based algorithm for single image de-raining. The basic idea is to sparsely approximate the patches of two layers by very high discriminative codes over a learned dictionary with strong mutual exclusivity property. Such discriminative sparse codes lead to accurate separation of two layers from their non-linear composite. The experiments showed that the proposed method outperformed the existing single image de-raining methods on tested rain images.

Many existing approaches for image-to-geometry registration assume that either a textured 3D model or a good initial guess of the 3D pose is available to bootstrap the registration process. In this paper we consider the registration of photographs to 3D models even when no texture information is available. This is very challenging as we cannot rely on texture gradients, and even shading gradients are hard to estimate since the lighting conditions are unknown. To that end, we propose average shading gradients, a rendering technique that estimates the average gradient magnitude over all lighting directions under Lambertian shading. We use this gradient representation as the building block of a registration pipeline based on matching sparse features. To cope with inevitable false matches due to the missing texture information and to increase robustness, the pose of the 3D model is estimated in two stages. Coarse pose hypotheses are first obtained from a single correct match each, subsequently refined using SIFT flow, and finally verified. We apply our algorithm to registering images of real-world objects to untextured 3D meshes of limited accuracy.

Coarse depth maps can be enhanced by using the shape information from polarization cues. We propose a framework to combine surface normals from polarization (hereafter polarization normals) with an aligned depth map. Polarization normals have not been used for depth enhancement before. This is because polarization normals suffer from physics-based artifacts, such as azimuthal ambiguity, refractive distortion and fronto-parallel signal degradation. We propose a framework to overcome these key challenges, allowing the benefits of polarization to be used to enhance depth maps. Our results demonstrate improvement with respect to state-of-the-art 3D reconstruction techniques.

Image segmentation is a critical step in many computer vision tasks, including high-level visual recognition and scene understanding as well as low-level photo and video processing. In this paper, we propose a new nonlinear embedding, called piecewise flat embedding, for image segmentation. Based on the theory of sparse signal recovery, piecewise flat embedding attempts to identify segment boundaries while significantly suppressing variations within segments. We adopt an L1-regularized energy term in the formulation to promote sparse solutions. We further devise an effective two-stage numerical algorithm based on Bregman iterations to solve the proposed embedding. Piecewise flat embedding can be easily integrated into existing image segmentation frameworks, including segmentation based on spectral clustering and hierarchical segmentation based on contour detection. Experiments on BSDS500 indicate that segmentation algorithms incorporating this embedding can achieve significantly improved results in both frameworks.

We present an approach to capture the 3D structure and motion of a group of people engaged in a social interaction. The core challenges in capturing social interactions are: (1) occlusion is functional and frequent; (2) subtle motion needs to be measured over a space large enough to host a social group; and (3) human appearance and configuration variation is immense. The Panoptic Studio is a system organized around the thesis that social interactions should be measured through the perceptual integration of a large variety of view points. We present a modularized system designed around this principle, consisting of integrated structural, hardware, and software innovations. The system takes, as input, 480 synchronized video streams of multiple people engaged in social activities, and produces, as output, the labeled time-varying 3D structure of anatomical landmarks on individuals in the space. The algorithmic contributions include a hierarchical approach for generating skeletal trajectory proposals, and an optimization framework for skeletal reconstruction with trajectory re-association.

We address the problem of hand pose estimation, formulated as an inverse problem. Typical approaches optimize an energy function over pose parameters using a `black box' image generation procedure. This procedure knows little about either the relationships between the parameters or the form of the energy function. In this paper, we show that we can significantly improving upon black box optimization by exploiting high-level knowledge of the structure of the parameters and using a local surrogate energy function. Our new framework, hierarchical sampling optimization, consists of a sequence of predictors organized into a kinematic hierarchy. Each predictor is conditioned on its ancestors, and generates a set of samples over a subset of the pose parameters. The highly-efficient surrogate energy is used to select among samples. Having evaluated the full hierarchy, the partial pose samples are concatenated to generate a full-pose hypothesis. Several hypotheses are generated using the same procedure, and finally the original full energy function selects the best result. Experimental evaluation on three publically available datasets show that our method is particularly impressive in low-compute scenarios where it significantly outperforms all other state-of-the-art methods.