ICCV 2019 Papers

In this paper we present a new computer vision task, named video instance segmentation. The goal of this new task is simultaneous detection, segmentation and tracking of instances in videos. In words, it is the first time that the image instance segmentation problem is extended to the video domain. To facilitate research on this new task, we propose a large-scale benchmark called YouTube-VIS, which consists of 2,883 high-resolution YouTube videos, a 40-category label set and 131k high-quality instance masks. In addition, we propose a novel algorithm called MaskTrack R-CNN for this task. Our new method introduces a new tracking branch to Mask R-CNN to jointly perform the detection, segmentation and tracking tasks simultaneously. Finally, we evaluate the proposed method and several strong baselines on our new dataset. Experimental results clearly demonstrate the advantages of the proposed algorithm and reveal insight for future improvement. We believe the video instance segmentation task will motivate the community along the line of research for video understanding.

Understanding videos such as TV series and movies requires analyzing who the characters are and what they are doing. We address the challenging problem of clustering face tracks based on their identity. Different from previous work in this area, we choose to operate in a realistic and difficult setting where: (i) the number of characters is not known a priori; and (ii) face tracks belonging to minor or background characters are not discarded. To this end, we propose Ball Cluster Learning (BCL), a supervised approach to carve the embedding space into balls of equal size, one for each cluster. The learned ball radius is easily translated to a stopping criterion for iterative merging algorithms. This gives BCL the ability to estimate the number of clusters as well as their assignment, achieving promising results on commonly used datasets. We also present a thorough discussion of how existing metric learning literature can be adapted for this task.

In this paper we present a a deep generative model for lossy video compression. We employ a model that consists of a 3D autoencoder with a discrete latent space and an autoregressive prior used for entropy coding. Both autoencoder and prior are trained jointly to minimize a rate-distortion loss, which is closely related to the ELBO used in variational autoencoders. Despite its simplicity, we find that our method outperforms the state-of-the-art learned video compression networks based on motion compensation or interpolation. We systematically evaluate various design choices, such as the use of frame-based or spatio-temporal autoencoders, and the type of autoregressive prior. In addition, we present three extensions of the basic method that demonstrate the benefits over classical approaches to compression. First, we introduce semantic compression, where the model is trained to allocate more bits to objects of interest. Second, we study adaptive compression, where the model is adapted to a domain with limited variability, e.g. videos taken from an autonomous car, to achieve superior compression on that domain. Finally, we introduce multimodal compression, where we demonstrate the effectiveness of our model in joint compression of multiple modalities captured by non-standard imaging sensors, such as quad cameras. We believe that this opens up novel video compression applications, which have not been feasible with classical codecs.

Group convolution has been shown to offer great computational savings in various 2D convolutional architectures for image classification. It is natural to ask: 1) if group convolution can help to alleviate the high computational cost of video classification networks; 2) what factors matter the most in 3D group convolutional networks; and 3) what are good computation/accuracy trade-offs with 3D group convolutional networks. This paper studies the effects of different design choices in 3D group convolutional networks for video classification. We empirically demonstrate that the amount of channel interactions plays an important role in the accuracy of 3D group convolutional networks. Our experiments suggest two main findings. First, it is a good practice to factorize 3D convolutions by separating channel interactions and spatiotemporal interactions as this leads to improved accuracy and lower computational cost. Second, 3D channel-separated convolutions provide a form of regularization, yielding lower training accuracy but higher test accuracy compared to 3D convolutions. These two empirical findings lead us to design an architecture -- Channel-Separated Convolutional Network (CSN) -- which is simple, efficient, yet accurate. On Sports1M and Kinetics, our CSNs are comparable with or better than the state-of-the-art while being 2-3 times more efficient.

In this paper we introduce ViSiL, a Video Similarity Learning architecture that considers fine-grained Spatio-Temporal relations between pairs of videos -- such relations are typically lost in previous video retrieval approaches that embed the whole frame or even the whole video into a vector descriptor before the similarity estimation. By contrast, our Convolutional Neural Network (CNN)-based approach is trained to calculate video-to-video similarity from refined frame-to-frame similarity matrices, so as to consider both intra- and inter-frame relations. In the proposed method, pairwise frame similarity is estimated by applying Tensor Dot (TD) followed by Chamfer Similarity (CS) on regional CNN frame features - this avoids feature aggregation before the similarity calculation between frames. Subsequently, the similarity matrix between all video frames is fed to a four-layer CNN, and then summarized using Chamfer Similarity (CS) into a video-to-video similarity score -- this avoids feature aggregation before the similarity calculation between videos and captures the temporal similarity patterns between matching frame sequences. We train the proposed network using a triplet loss scheme and evaluate it on five public benchmark datasets on four different video retrieval problems where we demonstrate large improvements in comparison to the state of the art. The implementation of ViSiL is publicly available.

We propose to learn word embeddings from visual co-occurrences. Two words co-occur visually if both words apply to the same image or image region. Specifically, we extract four types of visual co-occurrences between object and attribute words from large-scale, textually-annotated visual databases like VisualGenome and ImageNet. We then train a multi-task log-bilinear model that compactly encodes word "meanings" represented by each co-occurrence type into a single visual word-vector. Through unsupervised clustering, supervised partitioning, and a zero-shot-like generalization analysis we show that our word embeddings complement text-only embeddings like GloVe by better representing similarities and differences between visual concepts that are difficult to obtain from text corpora alone. We further evaluate our embeddings on five downstream applications, four of which are vision-language tasks. Augmenting GloVe with our embeddings yields gains on all tasks. We also find that random embeddings perform comparably to learned embeddings on all supervised vision-language tasks, contrary to conventional wisdom.

Comparing to image inpainting, image outpainting receives less attention due to two challenges in it. The first challenge is how to keep the spatial and content consistency between generated images and original input. The second challenge is how to maintain high quality in generated results, especially for multi-step generations in which generated regions are spatially far away from the initial input. To solve the two problems, we devise some innovative modules, named Skip Horizontal Connection and Recurrent Content Transfer, and integrate them into our designed encoder-decoder structure. By this design, our network can generate highly realistic outpainting prediction effectively and efficiently. Other than that, our method can generate new images with very long sizes while keeping the same style and semantic content as the given input. To test the effectiveness of the proposed architecture, we collect a new scenery dataset with diverse, complicated natural scenes. The experimental results on this dataset have demonstrated the efficacy of our proposed network.

In this work, we construct a large-scale dataset for vehicle re-identification (ReID), which contains 137k images of 13k vehicle instances captured by UAV-mounted cameras. To our knowledge, it is the largest UAV-based vehicle ReID dataset. To increase intra-class variation, each vehicle is captured by at least two UAVs at different locations, with diverse view-angles and flight-altitudes. We manually label a variety of vehicle attributes, including vehicle type, color, skylight, bumper, spare tire and luggage rack. Furthermore, for each vehicle image, the annotator is also required to mark the discriminative parts that helps them to distinguish this particular vehicle from others. Besides the dataset, we also design a specific vehicle ReID algorithm to make full use of the rich annotation information. It is capable of explicitly detecting discriminative parts for each specific vehicle and significantly outperforming the evaluated baselines and state-of-the-art vehicle ReID approaches.

This paper considers vehicle re-identification (re-ID) problem. The extreme viewpoint variation (up to 180 degrees) poses great challenges for existing approaches. Inspired by the behavior in human's recognition process, we propose a novel viewpoint-aware metric learning approach. It learns two metrics for similar viewpoints and different viewpoints in two feature spaces, respectively, giving rise to viewpoint-aware network (VANet). During training, two types of constraints are applied jointly. During inference, viewpoint is firstly estimated and the corresponding metric is used. Experimental results confirm that VANet significantly improves re-ID accuracy, especially when the pair is observed from different viewpoints. Our method establishes the new state-of-the-art on two benchmarks.

Photometric stereo (PS) techniques nowadays remain constrained to an ideal laboratory setup where modeling and calibration of lighting is amenable. To eliminate such restrictions, we propose an efficient principled variational approach to uncalibrated PS under general illumination. To this end, the Lambertian reflectance model is approximated through a spherical harmonic expansion, which preserves the spatial invariance of the lighting. The joint recovery of shape, reflectance and illumination is then formulated as a single variational problem. There the shape estimation is carried out directly in terms of the underlying perspective depth map, thus implicitly ensuring integrability and bypassing the need for a subsequent normal integration. To tackle the resulting nonconvex problem numerically, we undertake a two-phase procedure to initialize a balloon-like perspective depth map, followed by a "lagged" block coordinate descent scheme. The experiments validate efficiency and robustness of this approach. Across a variety of evaluations, we are able to reduce the mean angular error consistently by a factor of 2-3 compared to the state-of-the-art.

We propose a variational Bayesian framework for enhancing few-shot learning performance. This idea is motivated by the fact that single point based metric learning approaches are inherently noise-vulnerable and easy-to-be-biased. In a nutshell, stochastic variational inference is invoked to approximate bias-eliminated class specific sample distributions. In the meantime, a classifier-free prediction is attained by leveraging the distribution statistics on novel samples. Extensive experimental results on several benchmarks well demonstrate the effectiveness of our distribution-driven few-shot learning framework over previous point estimates based methods, in terms of superior classification accuracy and robustness.

Active learning aims to develop label-efficient algorithms by sampling the most representative queries to be labeled by an oracle. We describe a pool-based semi-supervised active learning algorithm that implicitly learns this sampling mechanism in an adversarial manner. Our method learns a latent space using a variational autoencoder (VAE) and an adversarial network trained to discriminate between unlabeled and labeled data. The mini-max game between the VAE and the adversarial network is played such that while the VAE tries to trick the adversarial network into predicting that all data points are from the labeled pool, the adversarial network learns how to discriminate between dissimilarities in the latent space. We extensively evaluate our method on various image classification and semantic segmentation benchmark datasets and establish a new state of the art on CIFAR10/100, Caltech-256, ImageNet, Cityscapes, and BDD100K. Our results demonstrate that our adversarial approach learns an effective low dimensional latent space in large-scale settings and provides for a computationally efficient sampling method. Our code is available at https://github.com/sinhasam/vaal.

In this paper, we propose a novel variable-rate learned image compression framework with a conditional autoencoder. Previous learning-based image compression methods mostly require training separate networks for different compression rates so they can yield compressed images of varying quality. In contrast, we train and deploy only one variable-rate image compression network implemented with a conditional autoencoder. We provide two rate control parameters, i.e., the Lagrange multiplier and the quantization bin size, which are given as conditioning variables to the network. Coarse rate adaptation to a target is performed by changing the Lagrange multiplier, while the rate can be further fine-tuned by adjusting the bin size used in quantizing the encoded representation. Our experimental results show that the proposed scheme provides a better rate-distortion trade-off than the traditional variable-rate image compression codecs such as JPEG2000 and BPG. Our model also shows comparable and sometimes better performance than the state-of-the-art learned image compression models that deploy multiple networks trained for varying rates.

We present a new large-scale multilingual video description dataset, VATEX, which contains over 41,250 videos and 825,000 captions in both English and Chinese. Among the captions, there are over 206,000 English-Chinese parallel translation pairs. Compared to the widely-used MSR-VTT dataset, \vatex is multilingual, larger, linguistically complex, and more diverse in terms of both video and natural language descriptions. We also introduce two tasks for video-and-language research based on \vatex: (1) Multilingual Video Captioning, aimed at describing a video in various languages with a compact unified captioning model, and (2) Video-guided Machine Translation, to translate a source language description into the target language using the video information as additional spatiotemporal context. Extensive experiments on the \vatex dataset show that, first, the unified multilingual model can not only produce both English and Chinese descriptions for a video more efficiently, but also offer improved performance over the monolingual models. Furthermore, we demonstrate that the spatiotemporal video context can be effectively utilized to align source and target languages and thus assist machine translation. In the end, we discuss the potentials of using \vatex for other video-and-language research.

We present a novel algorithm for point cloud segmentation.Our approach transforms unstructured point clouds into regular voxel grids, and further uses a kernel-based interpolated variational autoencoder (VAE) architecture to encode the local geometry within each voxel.Traditionally, the voxel representation only comprises Boolean occupancy information, which fails to capture the sparsely distributed points within voxels in a compact manner. In order to handle sparse distributions of points, we further employ radial basis functions (RBF) to compute a local, continuous representation within each voxel. Our approach results in a good volumetric representation that effectively tackles noisy point cloud datasets and is more robust for learning. Moreover, we further introduce group equivariant CNN to 3D, by defining the convolution operator on a symmetry group acting on Z ^3 and its isomorphic sets. This improves the expressive capacity without increasing parameters, leading to more robust segmentation results.We highlight the performance on standard benchmarks and show that our approach outperforms state-of-the-art segmentation algorithms on the ShapeNet and S3DIS datasets.

Image-based virtual try-on systems with the goal of transferring a desired clothing item onto the corresponding region of a person have made great strides recently, but challenges remain in generating realistic looking images that preserve both body and clothing details. Here we present a new virtual try-on network, called VTNFP, to synthesize photo-realistic images given the images of a clothed person and a target clothing item. In order to better preserve clothing and body features, VTNFP follows a three-stage design strategy. First, it transforms the target clothing into a warped form compatible with the pose of the given person. Next, it predicts a body segmentation map of the person wearing the target clothing, delineating body parts as well as clothing regions. Finally, the warped clothing, body segmentation map and given person image are fused together for fine-scale image synthesis. A key innovation of VTNFP is the body segmentation map prediction module, which provides critical information to guide image synthesis in regions where body parts and clothing intersects, and is very beneficial for preventing blurry pictures and preserving clothing and body part details. Experiments on a fashion dataset demonstrate that VTNFP generates substantially better results than state-of-the-art methods.

We introduce UprightNet, a learning-based approach for estimating 2DoF camera orientation from a single RGB image of an indoor scene. Unlike recent methods that leverage deep learning to perform black-box regression from image to orientation parameters, we propose an end-to-end framework that incorporates explicit geometric reasoning. In particular, we design a network that predicts two representations of scene geometry, in both the local camera and global reference coordinate systems, and solves for the camera orientation as the rotation that best aligns these two predictions via a differentiable least squares module. This network can be trained end-to-end, and can be supervised with both ground truth camera poses and intermediate representations of surface geometry. We evaluate UprightNet on the single-image camera orientation task on synthetic and real datasets, and show significant improvements over prior state-of-the-art approaches.

Learning to synthesize high frame rate videos via interpolation requires large quantities of high frame rate training videos, which, however, are scarce, especially at high resolutions. Here, we propose unsupervised techniques to synthesize high frame rate videos directly from low frame rate videos using cycle consistency. For a triplet of consecutive frames, we optimize models to minimize the discrepancy between the center frame and its cycle reconstruction, obtained by interpolating back from interpolated intermediate frames. This simple unsupervised constraint alone achieves results comparable with supervision using the ground truth intermediate frames. We further introduce a pseudo supervised loss term that enforces the interpolated frames to be consistent with predictions of a pre-trained interpolation model. The pseudo supervised loss term, used together with cycle consistency, can effectively adapt a pre-trained model to a new target domain. With no additional data and in a completely unsupervised fashion, our techniques significantly improve pre-trained models on new target domains, increasing PSNR values from 32.84dB to 33.05dB on the Slowflow and from 31.82dB to 32.53dB on the Sintel evaluation datasets.

Deep generative models come with the promise to learn an explainable representation for visual objects that allows image sampling, synthesis, and selective modification. The main challenge is to learn to properly model the independent latent characteristics of an object, especially its appearance and pose. We present a novel approach that learns disentangled representations of these characteristics and explains them individually. Training requires only pairs of images depicting the same object appearance, but no pose annotations. We propose an additional classifier that estimates the minimal amount of regularization required to enforce disentanglement. Thus both representations together can completely explain an image while being independent of each other. Previous methods based on adversarial approaches fail to enforce this independence, while methods based on variational approaches lead to uninformative representations. In experiments on diverse object categories, the approach successfully recombines pose and appearance to reconstruct and retarget novel synthesized images. We achieve significant improvements over state-of-the-art methods which utilize the same level of supervision, and reach performances comparable to those of pose-supervised approaches. However, we can handle the vast body of articulated object classes for which no pose models/annotations are available.