ICCV 2017 论文列表

We address the problem of synthesizing new video frames in an existing video, either in-between existing frames (interpolation), or subsequent to them (extrapolation). This problem is challenging because video appearance and motion can be highly complex. Traditional optical-flow-based solutions often fail where flow estimation is challenging, while newer neural-network-based methods that hallucinate pixel values directly often produce blurry results. We combine the advantages of these two methods by training a deep network that learns to synthesize video frames by flowing pixel values from existing ones, which we call deep voxel flow. Our method requires no human supervision, and any video can be used as training data by dropping, and then learning to predict, existing frames. The technique is efficient, and can be applied at any video resolution. We demonstrate that our method produces results that both quantitatively and qualitatively improve upon the state-of-the-art.

Automatically learning the structure of object categories remains an important open problem in computer vision. We propose a novel unsupervised approach that can discover and learn to detect landmarks in object categories, thus characterizing their structure. Our approach is based on factorizing image deformations, as induced by a viewpoint change or an object articulation, by learning a deep neural network that detects landmarks compatible with such visual effects. We show that, by requiring the same neural network to be applicable to different object instances, our method naturally induces meaningful correspondences between different object instances in a category. We assess the method qualitatively on a variety of object types, natural an man-made. We also show that our unsupervised landmarks are highly predictive of manually-annotated landmarks in faces benchmark datasets, and can be used to regress those with a high degree of accuracy.

We study the problem of mapping an input image to a tied pair consisting of a vector of parameters and an image that is created using a graphical engine from the vector of parameters. The mapping's objective is to have the output image as similar as possible to the input image. During training, no supervision is given in the form of matching inputs and outputs. This learning problem extends two literature problems: unsupervised domain adaptation and cross domain transfer. We define a generalization bound that is based on discrepancy, and employ a GAN to implement a network solution that corresponds to this bound. Experimentally, our method is shown to solve the problem of automatically creating avatars.

Despite the substantial progress in recent years, the problem of image captioning remains far from being satisfactorily tackled. Sentences produced by existing methods, e.g. those based on LSTM, are often overly rigid and lacking in variability. This issue is related to a learning principle widely used in practice, that is, to maximize the likelihood of training samples. This principle encourages the high resemblance to the "ground-truths", while suppressing other reasonable expressions. Conventional evaluation metrics, e.g. BLEU and METEOR, also favor such restrictive methods. In this paper, we explore an alternative approach, with an aim to improve the naturalness and diversity - two essential properties of human expressions. Specifically, we propose a new framework based on Conditional Generative Adversarial Networks (CGAN), which jointly learns a generator to produce descriptions conditioned on images and an evaluator to assess how well a description fits the visual content. It is noteworthy that training a sequence generator is nontrivial. We overcome the difficulty by Policy Gradient, a strategy stemming from Reinforcement Learning, which allows the generator to receive early feedbacks along the way. We tested our method on two large datasets, where it performed competitively against real people in our user study and outperformed other methods on various tasks.

Modeling the long-term facial aging process is extremely challenging due to the presence of large and non-linear variations during the face development stages. In order to efficiently address the problem, this work first decomposes the aging process into multiple short-term stages. Then, a novel generative probabilistic model, named Temporal Non-Volume Preserving (TNVP) transformation, is presented to model the facial aging process at each stage. Unlike Generative Adversarial Networks (GANs), which requires an empirical balance threshold, and Restricted Boltzmann Machines (RBM), an intractable model, our proposed TNVP approach guarantees a tractable density function, exact inference and evaluation for embedding the feature transformations between faces in consecutive stages. Our model shows its advantages not only in capturing the non-linear age related variance in each stage but also producing a smooth synthesis in age progression across faces. Our approach can model any face in the wild provided with only four basic landmark points. Moreover, the structure can be transformed into a deep convolutional network while keeping the advantages of probabilistic models with tractable log-likelihood density estimation. Our method is evaluated in both terms of synthesizing age-progressed faces and cross-age face verification and consistently shows the state-of-the-art results in various face aging databases, i.e. FG-NET, MORPH, AginG Faces in the Wild (AGFW), and Cross-Age Celebrity Dataset (CACD). A large-scale face verification on Megaface challenge 1 is also performed to further show the advantages of our proposed approach.

Synthesizing high-quality images from text descriptions is a challenging problem in computer vision and has many practical applications. Samples generated by existing text-to-image approaches can roughly reflect the meaning of the given descriptions, but they fail to contain necessary details and vivid object parts. In this paper, we propose Stacked Generative Adversarial Networks (StackGAN) to generate 256x256 photo-realistic images conditioned on text descriptions. We decompose the hard problem into more manageable sub-problems through a sketch-refinement process. The Stage-I GAN sketches the primitive shape and colors of the object based on the given text description, yielding Stage-I low-resolution images. The Stage-II GAN takes Stage-I results and text descriptions as inputs, and generates high-resolution images with photo-realistic details. It is able to rectify defects in Stage-I results and add compelling details with the refinement process. To improve the diversity of the synthesized images and stabilize the training of the conditional-GAN, we introduce a novel Conditioning Augmentation technique that encourages smoothness in the latent conditioning manifold. Extensive experiments and comparisons with state-of-the-arts on benchmark datasets demonstrate that the proposed method achieves significant improvements on generating photo-realistic images conditioned on text descriptions.

We aim for zero-shot localization and classification of human actions in video. Where traditional approaches rely on global attribute or object classification scores for their zero-shot knowledge transfer, our main contribution is a spatial-aware object embedding. To arrive at spatial awareness, we build our embedding on top of freely available actor and object detectors. Relevance of objects is determined in a word embedding space and further enforced with estimated spatial preferences. Besides local object awareness, we also embed global object awareness into our embedding to maximize actor and object interaction. Finally, we exploit the object positions and sizes in the spatial-aware embedding to demonstrate a new spatio-temporal action retrieval scenario with composite queries. Action localization and classification experiments on four contemporary action video datasets support our proposal. Apart from state-of-the-art results in the zero-shot localization and classification settings, our spatial-aware embedding is even competitive with recent supervised action localization alternatives.

We introduce scGAN, a novel extension of conditional Generative Adversarial Networks (GAN) tailored for the challenging problem of shadow detection in images. Previous methods for shadow detection focus on learning the local appearance of shadow regions, while using limited local context reasoning in the form of pairwise potentials in a Conditional Random Field. In contrast, the proposed adversarial approach is able to model higher level relationships and global scene characteristics. We train a shadow detector that corresponds to the generator of a conditional GAN, and augment its shadow accuracy by combining the typical GAN loss with a data loss term. Due to the unbalanced distribution of the shadow labels, we use weighted cross entropy. With the standard GAN architecture, properly setting the weight for the cross entropy would require training multiple GANs, a computationally expensive grid procedure. In scGAN, we introduce an additional sensitivity parameter w to the generator. The proposed approach effectively parameterizes the loss of the trained detector. The resulting shadow detector is a single network that can generate shadow maps corresponding to different sensitivity levels, obviating the need for multiple models and a costly training procedure. We evaluate our method on the large-scale SBU and UCF shadow datasets, and observe up to 17% error reduction with respect to the previous state-of-the-art method.

In this work, we propose a technique to convert CNN models for semantic segmentation of static images into CNNs for video data. We describe a warping method that can be used to augment existing architectures with very little extra computational cost. This module is called NetWarp and we demonstrate its use for a range of network architectures. The main design principle is to use optical flow of adjacent frames for warping internal network representations across time. A key insight of this work is that fast optical flow methods can be combined with many different CNN architectures for improved performance and end-to-end training. Experiments validate that the proposed approach incurs only little extra computational cost, while improving performance, when video streams are available. We achieve new state-of-the-art results on the CamVid and Cityscapes benchmark datasets and show consistent improvements over different baseline networks. Our code and models are available at http://segmentation.is.tue.mpg.de

We present a novel method for detecting 3D model instances and estimating their 6D poses from RGB data in a single shot. To this end, we extend the popular SSD paradigm to cover the full 6D pose space and train on synthetic model data only. Our approach competes or surpasses current state-of-the-art methods that leverage RGB-D data on multiple challenging datasets. Furthermore, our method produces these results at around 10Hz, which is many times faster than the related methods. For the sake of reproducibility, we make our trained networks and detection code publicly available.

Markov Random Fields are widely used to model light-field stereo matching problems. However, most previous approaches used fixed parameters and did not adapt to light-field statistics. Instead, they explored explicit vision cues to provide local adaptability and thus enhanced depth quality. But such additional assumptions could end up confining their applicability, e.g. algorithms designed for dense light fields are not suitable for sparse ones. In this paper, we develop an empirical Bayesian framework--Robust Pseudo Random Field--to explore intrinsic statistical cues for broad applicability. Based on pseudo-likelihood, it applies soft expectation-maximization (EM) for good model fitting and hard EM for robust depth estimation. We introduce novel pixel difference models to enable such adaptability and robustness simultaneously. We also devise an algorithm to employ this framework on dense, sparse, and even denoised light fields. Experimental results show that it estimates scene-dependent parameters robustly and converges quickly. In terms of depth accuracy and computation speed, it also outperforms state-of-the-art algorithms constantly.

We introduce a novel method for representation learning that uses an artificial supervision signal based on counting visual primitives. This supervision signal is obtained from an equivariance relation, which does not require any manual annotation. We relate transformations of images to transformations of the representations. More specifically, we look for the representation that satisfies such relation rather than the transformations that match a given representation. In this paper, we use two image transformations in the context of counting: scaling and tiling. The first transformation exploits the fact that the number of visual primitives should be invariant to scale. The second transformation allows us to equate the total number of visual primitives in each tile to that in the whole image. These two transformations are combined in one constraint and used to train a neural network with a contrastive loss. The proposed task produces representations that perform on par or exceed the state of the art in transfer learning benchmarks.

Recent studies demonstrate the effectiveness of Recurrent Neural Networks (RNNs) for action recognition in videos. However, previous works mainly utilize video-level category as supervision to train RNNs, which may prohibit RNNs to learn complex motion structures along time. In this paper, we propose a recurrent pose-attention network (RPAN) to address this challenge, where we introduce a novel pose-attention mechanism to adaptively learn pose-related features at every time-step action prediction of RNNs. More specifically, we make three main contributions in this paper. Firstly, unlike previous works on pose-related action recognition, our RPAN is an end-to-end recurrent network which can exploit important spatial-temporal evolutions of human pose to assist action recognition in a unified framework. Secondly, instead of learning individual human-joint features separately, our pose-attention mechanism learns robust human-part features by sharing attention parameters partially on the semantically-related human joints. These human-part features are then fed into the human-part pooling layer to construct a highly-discriminative pose-related representation for temporal action modeling. Thirdly, one important byproduct of our RPAN is pose estimation in videos, which can be used for coarse pose annotation in action videos. We evaluate the proposed RPAN quantitatively and qualitatively on two popular benchmarks, i.e., Sub-JHMDB and PennAction. Experimental results show that RPAN outperforms the recent state-of-the-art methods on these challenging datasets.

This paper presents a solution to the Projective Structure from Motion (PSfM) problem able to deal efficiently with missing data, outliers and, for the first time, large scale 3D reconstruction scenarios. By embedding the projective depths into the projective parameters of the points and views, we decrease the number of unknowns to estimate and improve computational speed by optimizing standard linear Least Squares systems instead of homogeneous ones. In order to do so, we show that an extension of the linear constraints from the Generalized Projective Reconstruction Theorem can be transferred to the projective parameters, ensuring also a valid projective reconstruction in the process. We use an incremental approach that, starting from a solvable sub-problem, incrementally adds views and points until completion with a robust, outliers free, procedure. Experiments with simulated data shows that our approach is performing well, both in term of the quality of the reconstruction and the capacity to handle missing data and outliers with a reduced computational time. Finally, results on real datasets shows the ability of the method to be used in medium and large scale 3D reconstruction scenarios with high ratios of missing data (up to 98%).

We present an approach to synthesizing photographic images conditioned on semantic layouts. Given a semantic label map, our approach produces an image with photographic appearance that conforms to the input layout. The approach thus functions as a rendering engine that takes a two-dimensional semantic specification of the scene and produces a corresponding photographic image. Unlike recent and contemporaneous work, our approach does not rely on adversarial training. We show that photographic images can be synthesized from semantic layouts by a single feedforward network with appropriate structure, trained end-to-end with a direct regression objective. The presented approach scales seamlessly to high resolutions; we demonstrate this by synthesizing photographic images at 2-megapixel resolution, the full resolution of our training data. Extensive perceptual experiments on datasets of outdoor and indoor scenes demonstrate that images synthesized by the presented approach are considerably more realistic than alternative approaches.

When the training and the test data belong to different domains, the accuracy of an object classifier is significantly reduced. Therefore, several algorithms have been proposed in the last years to diminish the so called domain shift between datasets. However, all available evaluation protocols for domain adaptation describe a closed set recognition task, where both domains, namely source and target, contain exactly the same object classes. In this work, we also explore the field of domain adaptation in open sets, which is a more realistic scenario where only a few categories of interest are shared between source and target data. Therefore, we propose a method that fits in both closed and open set scenarios. The approach learns a mapping from the source to the target domain by jointly solving an assignment problem that labels those target instances that potentially belong to the categories of interest present in the source dataset. A thorough evaluation shows that our approach outperforms the state-of-the-art.

While deep learning methods have achieved state-of-the-art performance in many challenging inverse problems like image inpainting and super-resolution, they invariably involve problem-specific training of the networks. Under this approach, each inverse problem requires its own dedicated network. In scenarios where we need to solve a wide variety of problems, e.g., on a mobile camera, it is inefficient and expensive to use these problem-specific networks. On the other hand, traditional methods using analytic signal priors can be used to solve any linear inverse problem; this often comes with a performance that is worse than learning-based methods. In this work, we provide a middle ground between the two kinds of methods -- we propose a general framework to train a single deep neural network that solves arbitrary linear inverse problems. We achieve this by training a network that acts as a quasi-projection operator for the set of natural images and show that any linear inverse problem involving natural images can be solved using iterative methods. We empirically show that the proposed framework demonstrates superior performance over traditional methods using wavelet sparsity prior while achieving performance comparable to specially-trained networks on tasks including compressive sensing and pixel-wise inpainting.

In this work we propose a novel model-based deep convolutional autoencoder that addresses the highly challenging problem of reconstructing a 3D human face from a single in-the-wild color image. To this end, we combine a convolutional encoder network with an expert-designed generative model that serves as decoder. The core innovation is the differentiable parametric decoder that encapsulates image formation analytically based on a generative model. Our decoder takes as input a code vector with exactly defined semantic meaning that encodes detailed face pose, shape, expression, skin reflectance and scene illumination. Due to this new way of combining CNN-based with model-based face reconstruction, the CNN-based encoder learns to extract semantically meaningful parameters from a single monocular input image. For the first time, a CNN encoder and an expert-designed generative model can be trained end-to-end in an unsupervised manner, which renders training on very large (unlabeled) real world data feasible. The obtained reconstructions compare favorably to current state-of-the-art approaches in terms of quality and richness of representation.

We present a conceptually simple, flexible, and general framework for object instance segmentation. Our approach efficiently detects objects in an image while simultaneously generating a high-quality segmentation mask for each instance. The method, called Mask R-CNN, extends Faster R-CNN by adding a branch for predicting an object mask in parallel with the existing branch for bounding box recognition. Mask R-CNN is simple to train and adds only a small overhead to Faster R-CNN, running at 5 fps. Moreover, Mask R-CNN is easy to generalize to other tasks, e.g., allowing us to estimate human poses in the same framework. We show top results in all three tracks of the COCO suite of challenges, including instance segmentation, bounding-box object detection, and person keypoint detection. Without tricks, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners. We hope our simple and effective approach will serve as a solid baseline and help ease future research in instance-level recognition. Code will be made available.