CVPR 2018 Papers

Standard video encoders developed for conventional narrow field-of-view video are widely applied to 360° video as well, with reasonable results. However, while this approach commits arbitrarily to a projection of the spherical frames, we observe that some orientations of a 360° video, once projected, are more compressible than others. We introduce an approach to predict the sphere rotation that will yield the maximal compression rate. Given video clips in their original encoding, a convolutional neural network learns the association between a clip’s visual content and its compressibility at different rotations of a cubemap projection. Given a novel video, our learning-based approach efficiently infers the most compressible direction in one shot, without repeated rendering and compression of the source video. We validate our idea on thousands of video clips and multiple popular video codecs. The results show that this untapped dimension of 360° compression has substantial potential—“good” rotations are typically 8−10% more compressible than bad ones, and our learning approach can predict them reliably 82% of the time.

Compositionality of semantic concepts in image synthesis and analysis is appealing as it can help in decomposing known and generatively recomposing unknown data. For instance, we may learn concepts of changing illumination, geometry or albedo of a scene, and try to recombine them to generate physically meaningful, but unseen data for training and testing. In practice however we often do not have samples from the joint concept space available: We may have data on illumination change in one data set and on geometric change in another one without complete overlap. We pose the following question: How can we learn two or more concepts jointly from different data sets with mutual consistency where we do not have samples from the full joint space? We present a novel answer in this paper based on cyclic consistency over multiple concepts, represented individually by generative adversarial networks (GANs). Our method, ConceptGAN, can be understood as a drop in for data augmentation to improve resilience for real world applications. Qualitative and quantitative evaluations demonstrate its efficacy in generating semantically meaningful images, as well as one shot face verification as an example application.

Recurrent Neural Networks (RNNs) are powerful sequence modeling tools. However, when dealing with high dimensional inputs, the training of RNNs becomes computational expensive due to the large number of model parameters. This hinders RNNs from solving many important computer vision tasks, such as Action Recognition in Videos and Image Captioning. To overcome this problem, we propose a compact and flexible structure, namely Block-Term tensor decomposition, which greatly reduces the parameters of RNNs and improves their training efficiency. Compared with alternative low-rank approximations, such as tensor-train RNN (TT-RNN), our method, Block-Term RNN (BT-RNN), is not only more concise (when using the same rank), but also able to attain a better approximation to the original RNNs with much fewer parameters. On three challenging tasks, including Action Recognition in Videos, Image Captioning and Image Generation, BT-RNN outperforms TT-RNN and the standard RNN in terms of both prediction accuracy and convergence rate. Specifically, BT-LSTM utilizes 17,388 times fewer parameters than the standard LSTM to achieve an accuracy improvement over 15.6% in the Action Recognition task on the UCF11 dataset.

In this paper, we investigate ways of conducting a detailed fashion search using query images and attributes. A credible fashion search platform should be able to (1) find images that share the same attributes as the query image, (2) allow users to manipulate certain attributes, e.g. replace collar attribute from round to v-neck, and (3) handle region-specific attribute manipulations, e.g. replacing the color attribute of the sleeve region without changing the color attribute of other regions. A key challenge to be addressed is that fashion products have multiple attributes and it is important for each of these attributes to have representative features. To address these challenges, we propose the FashionSearchNet which uses a weakly supervised localization method to extract regions of attributes. By doing so, unrelated features can be ignored thus improving the similarity learning. Also, FashionSearchNet incorporates a new procedure that enables region awareness to be able to handle region-specific requests. FashionSearchNet outperforms the most recent fashion search techniques and is shown to be able to carry out different search scenarios using the dynamic queries.

Offline training for object tracking has recently shown great potentials in balancing tracking accuracy and speed. However, it is still difficult to adapt an offline trained model to a target tracked online. This work presents a Residual Attentional Siamese Network (RASNet) for high performance object tracking. The RASNet model reformulates the correlation filter within a Siamese tracking framework, and introduces different kinds of the attention mechanisms to adapt the model without updating the model online. In particular, by exploiting the offline trained general attention, the target adapted residual attention, and the channel favored feature attention, the RASNet not only mitigates the over-fitting problem in deep network training, but also enhances its discriminative capacity and adaptability due to the separation of representation learning and discriminator learning. The proposed deep architecture is trained from end to end and takes full advantage of the rich spatial temporal information to achieve robust visual tracking. Experimental results on two latest benchmarks, OTB-2015 and VOT2017, show that the RASNet tracker has the state-of-the-art tracking accuracy while runs at more than 80 frames per second.

We propose a novel probabilistic model for visual question answering (Visual QA). The key idea is to infer two sets of embeddings: one for the image and the question jointly and the other for the answers. The learning objective is to learn the best parameterization of those embeddings such that the correct answer has higher likelihood among all possible answers. In contrast to several existing approaches of treating Visual QA as multi-way classification, the proposed approach takes the semantic relationships (as characterized by the embeddings) among answers into consideration, instead of viewing them as independent ordinal numbers. Thus, the learned embedded function can be used to embed unseen answers (in the training dataset). These properties make the approach particularly appealing for transfer learning for open-ended Visual QA, where the source dataset on which the model is learned has limited overlapping with the target dataset in the space of answers. We have also developed large-scale optimization techniques for applying the model to datasets with a large number of answers, where the challenge is to properly normalize the proposed probabilistic models. We validate our approach on several Visual QA datasets and investigate its utility for transferring models across datasets. The empirical results have shown that the approach performs well not only on in-domain learning but also on transfer learning.

3D shape completion from partial point clouds is a fundamental problem in computer vision and computer graphics. Recent approaches can be characterized as either data-driven or learning-based. Data-driven approaches rely on a shape model whose parameters are optimized to fit the observations. Learning-based approaches, in contrast, avoid the expensive optimization step and instead directly predict the complete shape from the incomplete observations using deep neural networks. However, full supervision is required which is often not available in practice. In this work, we propose a weakly-supervised learning-based approach to 3D shape completion which neither requires slow optimization nor direct supervision. While we also learn a shape prior on synthetic data, we amortize, i.e., learn, maximum likelihood fitting using deep neural networks resulting in efficient shape completion without sacrificing accuracy. Tackling 3D shape completion of cars on ShapeNet and KITTI, we demonstrate that the proposed amortized maximum likelihood approach is able to compete with a fully supervised baseline and a state-of-the-art data-driven approach while being significantly faster. On ModelNet, we additionally show that the approach is able to generalize to other object categories as well.

We address the problem of texture segmentation by grouping dense pixel-wise descriptors. We introduce and construct learned Shape-Tailored Descriptors that aggregate image statistics only within regions of interest to avoid mixing statistics of different textures, and that are invariant to complex nuisances (e.g., illumination, perspective and deformations). This is accomplished by training a neural network to discriminate base shape-tailored descriptors of oriented gradients at various scales. These descriptors are defined through partial differential equations to obtain data at various scales in arbitrarily shaped regions. We formulate and optimize a joint optimization problem in the segmentation and descriptors to discriminate these base descriptors using the learned metric, equivalent to grouping learned descriptors. We test the method on datasets to illustrate the effect of both the shape-tailored and learned properties of the descriptors. Experiments show that the descriptors learned on a small dataset of segmented images generalize well to unseen textures in other datasets, showing the generic nature of these descriptors. We show stateof- the-art results on texture segmentation benchmarks.

We introduce a method for efficiently crowdsourcing multiclass annotations in challenging, real world image datasets. Our method is designed to minimize the number of human annotations that are necessary to achieve a desired level of confidence on class labels. It is based on combining models of worker behavior with computer vision. Our method is general: it can handle a large number of classes, worker labels that come from a taxonomy rather than a flat list, and can model the dependence of labels when workers can see a history of previous annotations. Our method may be used as a drop-in replacement for the majority vote algorithms used in online crowdsourcing services that aggregate multiple human annotations into a final consolidated label. In experiments conducted on two real-life applications we find that our method can reduce the number of required annotations by as much as a factor of 5.4 and can reduce the residual annotation error by up to 90% when compared with majority voting. Furthermore, the online risk estimates of the models may be used to sort the annotated collection and minimize subsequent expert review effort.

We propose an algorithm to predict room layout from a single image that generalizes across panoramas and perspective images, cuboid layouts and more general layouts (e.g. "L"-shape room). Our method operates directly on the panoramic image, rather than decomposing into perspective images as do recent works. Our network architecture is similar to that of RoomNet, but we show improvements due to aligning the image based on vanishing points, predicting multiple layout elements (corners, boundaries, size and translation), and fitting a constrained Manhattan layout to the resulting predictions. Our method compares well in speed and accuracy to other existing work on panoramas, achieves among the best accuracy for perspective images, and can handle both cuboid-shaped and more general Manhattan layouts.

We present a method that can evaluate a RANSAC hypothesis in constant time, i.e. independent of the size of the data. A key observation here is that correct hypotheses are tightly clustered together in the latent parameter domain. In a manner similar to the generalized Hough transform we seek to find this cluster, only that we need as few as two votes for a successful detection. Rapidly locating such pairs of similar hypotheses is made possible by adapting the recent "Random Grids" range-search technique. We only perform the usual (costly) hypothesis verification stage upon the discovery of a close pair of hypotheses. We show that this event rarely happens for incorrect hypotheses, enabling a significant speedup of the RANSAC pipeline. The suggested approach is applied and tested on three robust estimation problems: camera localization, 3D rigid alignment and 2D-homography estimation. We perform rigorous testing on both synthetic and real datasets, demonstrating an improvement in efficiency without a compromise in accuracy. Furthermore, we achieve state-of-the-art 3D alignment results on the challenging ``Redwood'' loop-closure challenge.

We propose a novel deep learning-based framework to tackle the challenge of semantic segmentation of large-scale point clouds of millions of points. We argue that the organization of 3D point clouds can be efficiently captured by a structure called superpoint graph (SPG), derived from a partition of the scanned scene into geometrically homogeneous elements. SPGs offer a compact yet rich representation of contextual relationships between object parts, which is then exploited by a graph convolutional network. Our framework sets a new state of the art for segmenting outdoor LiDAR scans (+11.9 and +8.8 mIoU points for both Semantic3D test sets), as well as indoor scans (+12.4 mIoU points for the S3DIS dataset).

Distance metric learning (DML) has been studied extensively in the past decades for its superior performance with distance-based algorithms. Most of the existing methods propose to learn a distance metric with pairwise or triplet constraints. However, the number of constraints is quadratic or even cubic in the number of the original examples, which makes it challenging for DML to handle the large-scale data set. Besides, the real-world data may contain various uncertainty, especially for the image data. The uncertainty can mislead the learning procedure and cause the performance degradation. By investigating the image data, we find that the original data can be observed from a small set of clean latent examples with different distortions. In this work, we propose the margin preserving metric learning framework to learn the distance metric and latent examples simultaneously. By leveraging the ideal properties of latent examples, the training efficiency can be improved significantly while the learned metric also becomes robust to the uncertainty in the original data. Furthermore, we can show that the metric is learned from latent examples only, but it can preserve the large margin property even for the original data. The empirical study on the benchmark image data sets demonstrates the efficacy and efficiency of the proposed method.

Transferring the knowledge learned from large scale datasets (e.g., ImageNet) via fine-tuning offers an effective solution for domain-specific fine-grained visual categorization (FGVC) tasks (e.g., recognizing bird species or car make & model). In such scenarios, data annotation often calls for specialized domain knowledge and thus is difficult to scale. In this work, we first tackle a problem in large scale FGVC. Our method won first place in iNaturalist 2017 large scale species classification challenge. Central to the success of our approach is a training scheme that uses higher image resolution and deals with the long-tailed distribution of training data. Next, we study transfer learning via fine-tuning from large scale datasets to small scale, domain-specific FGVC datasets. We propose a measure to estimate domain similarity via Earth Mover's Distance and demonstrate that transfer learning benefits from pre-training on a source domain that is similar to the target domain by this measure. Our proposed transfer learning outperforms ImageNet pre-training and obtains state-of-the-art results on multiple commonly used FGVC datasets.

We investigate the problem of Language-Based Image Editing (LBIE). Given a source image and a natural language description, we want to generate a target image by editing the source image based on the description. We propose a generic modeling framework for two sub-tasks of LBIE: language-based image segmentation and image colorization. The framework uses recurrent attentive models to fuse image and language features. Instead of using a fixed step size, we introduce for each region of the image a termination gate to dynamically determine after each inference step whether to continue extrapolating additional information from the textual description. The effectiveness of the framework is validated on three datasets. First, we introduce a synthetic dataset, called CoSaL, to evaluate the end-to-end performance of our LBIE system. Second, we show that the framework leads to state-of-the-art performance on image segmentation on the ReferIt dataset. Third, we present the first language-based colorization result on the Oxford-102 Flowers dataset.

Lighting estimation from faces is an important task and has applications in many areas such as image editing, intrinsic image decomposition, and image forgery detection. We propose to train a deep Convolutional Neural Network (CNN) to regress lighting parameters from a single face image. Lacking massive ground truth lighting labels for face images in the wild, we use an existing method to estimate lighting parameters, which are treated as ground truth with noise. To alleviate the effect of such noise, we utilize the idea of Generative Adversarial Networks (GAN) and propose a Label Denoising Adversarial Network (LDAN). LDAN makes use of synthetic data with accurate ground truth to help train a deep CNN for lighting regression on real face images. Experiments show that our network outperforms existing methods in producing consistent lighting parameters of different faces under similar lighting conditions. To further evaluate the proposed method, we also apply it to regress object 2D key points where ground truth labels are available. Our experiments demonstrate its effectiveness on this application.

We observed that recent state-of-the-art results on single image human pose estimation were achieved by multi-stage Convolution Neural Networks (CNN). Notwithstanding the superior performance on static images, the application of these models on videos is not only computationally intensive, it also suffers from performance degeneration and flicking. Such suboptimal results are mainly attributed to the inability of imposing sequential geometric consistency, handling severe image quality degradation (e.g. motion blur and occlusion) as well as the inability of capturing the temporal correlation among video frames. In this paper, we proposed a novel recurrent network to tackle these problems. We showed that if we were to impose the weight sharing scheme to the multi-stage CNN, it could be re-written as a Recurrent Neural Network (RNN). This property decouples the relationship among multiple network stages and results in significantly faster speed in invoking the network for videos. It also enables the adoption of Long Short-Term Memory (LSTM) units between video frames. We found such memory augmented RNN is very effective in imposing geometric consistency among frames. It also well handles input quality degradation in videos while successfully stabilizes the sequential outputs. The experiments showed that our approach significantly outperformed current state-of-the-art methods on two large-scale video pose estimation benchmarks. We also explored the memory cells inside the LSTM and provided insights on why such mechanism would benefit the prediction for video-based pose estimations.

We present the first end-to-end approach for real-time material estimation for general object shapes with uniform material that only requires a single color image as input. In addition to Lambertian surface properties, our approach fully automatically computes the specular albedo, material shininess, and a foreground segmentation. We tackle this challenging and ill-posed inverse rendering problem using recent advances in image-to-image translation techniques based on deep convolutional encoder–decoder architectures. The underlying core representations of our approach are specular shading, diffuse shading and mirror images, which allow to learn the effective and accurate separation of diffuse and specular albedo. In addition, we propose a novel highly efficient perceptual rendering loss that mimics real world image formation and obtains intermediate results even during run time. The estimation of material parameters at real-time frame rates enables exciting mixed reality applications, such as seamless illumination-consistent integration of virtual objects into realworld scenes, and virtual material cloning.We demonstrate our approach in a live setup, compare it to the state of the art, and demonstrate its effectiveness through quantitative and qualitative evaluation.

Learning to estimate 3D geometry in a single image by watching unlabeled videos via deep convolutional network is attracting significant attention. In this paper, we introduce a “3D as-smooth-as-possible (3D-ASAP)” prior inside the pipeline, which enables joint estimation of edges and 3D scene, yielding results with significant improvement in accuracy for fine detailed structures. Specifically, we define the 3D-ASAP prior by requiring that any two points recovered in 3D from an image should lie on an existing planar surface if no other cues provided. We design an unsupervised framework that Learns Edges and Geometry (depth, normal) all at Once (LEGO). The predicted edges are embedded into depth and surface normal smoothness terms, where pixels without edges in-between are constrained to satisfy the prior. In our framework, the predicted depths, normals and edges are forced to be consistent all the time. We conduct experiments on KITTI to evaluate our estimated geometry and CityScapes to perform edge evaluation. We show that in all of the tasks, i.e. depth, normal and edge, our algorithm vastly outperforms other state-of-the-art (SOTA) algorithms, demonstrating the benefits of our approach.