CVPR 2017 Papers

Human age is considered an important biometric trait for human identification or search. Recent research shows that the aging features deeply learned from large-scale data lead to significant performance improvement on facial image-based age estimation. However, age-related ordinal information is totally ignored in these approaches. In this paper, we propose a novel Convolutional Neural Network (CNN)-based framework, ranking-CNN, for age estimation. Ranking-CNN contains a series of basic CNNs, each of which is trained with ordinal age labels. Then, their binary outputs are aggregated for the final age prediction. We theoretically obtain a much tighter error bound for ranking-based age estimation. Moreover, we rigorously prove that ranking-CNN is more likely to get smaller estimation errors when compared with multi-class classification approaches. Through extensive experiments, we show that statistically, ranking-CNN significantly outperforms other state-of-the-art age estimation models on benchmark datasets.

We investigate the problem of estimating the dense 3D shape of an object, given a set of 2D landmarks and silhouette in a single image. An obvious prior to employ in such a problem is a dictionary of dense CAD models. Employing a sufficiently large enough dictionary of CAD models, however, is in general computationally infeasible. A common strategy in dictionary learning to encourage generalization is to allow for linear combinations of dictionary elements. This too, however, is problematic as most CAD models cannot be readily placed in global dense correspondence. In this paper, we propose a two-step strategy. First, we employ orthogonal matching pursuit to rapidly choose the "closest" single CAD model in our dictionary to the projected image. Second, we employ a novel graph embedding based on local dense correspondence to allow for sparse linear combinations of CAD models. We validate our framework experimentally in both synthetic and real world scenario and demonstrate the superiority of our approach to both 3D mesh reconstruction and volumetric representation.

Current action recognition methods heavily rely on trimmed videos for model training. However, it is expensive and time-consuming to acquire a large-scale trimmed video dataset. This paper presents a new weakly supervised architecture, called UntrimmedNet, which is able to directly learn action recognition models from untrimmed videos without the requirement of temporal annotations of action instances. Our UntrimmedNet couples two important components, the classification module and the selection module, to learn the action models and reason about the temporal duration of action instances, respectively. These two components are implemented with feed-forward networks, and UntrimmedNet is therefore an end-to-end trainable architecture. We exploit the learned models for action recognition (WSR) and detection (WSD) on the untrimmed video datasets of THUMOS14 and ActivityNet. Although our UntrimmedNet only employs weak supervision, our method achieves performance superior or comparable to that of those strongly supervised approaches on these two datasets.

We propose an unsupervised method for reference resolution in instructional videos, where the goal is to temporally link an entity (e.g., "dressing") to the action (e.g., "mix yogurt") that produced it. The key challenge is the inevitable visual-linguistic ambiguities arising from the changes in both visual appearance and referring expression of an entity in the video. This challenge is amplified by the fact that we aim to resolve references with no supervision. We address these challenges by learning a joint visual-linguistic model, where linguistic cues can help resolve visual ambiguities and vice versa. We verify our approach by learning our model unsupervisedly using more than two thousand unstructured cooking videos from YouTube, and show that our visual-linguistic model can substantially improve upon state-of-the-art linguistic only model on reference resolution in instructional videos.

This paper addresses the problem of unsupervised video summarization, formulated as selecting a sparse subset of video frames that optimally represent the input video. Our key idea is to learn a deep summarizer network to minimize distance between training videos and a distribution of their summarizations, in an unsupervised way. Such a summarizer can then be applied on a new video for estimating its optimal summarization. For learning, we specify a novel generative adversarial framework, consisting of the summarizer and discriminator. The summarizer is the autoencoder long short-term memory network (LSTM) aimed at, first, selecting video frames, and then decoding the obtained summarization for reconstructing the input video. The discriminator is another LSTM aimed at distinguishing between the original video and its reconstruction from the summarizer. The summarizer LSTM is cast as an adversary of the discriminator, i.e., trained so as to maximally confuse the discriminator. This learning is also regularized for sparsity. Evaluation on four benchmark datasets, consisting of videos showing diverse events in first- and third-person views, demonstrates our competitive performance in comparison to fully supervised state-of-the-art approaches.

The article concerns the automatic calibration of a camera with radial distortion from a single image. It is known that, under the mild assumption of square pixels and zero skew, lines in the scene project into circles in the image, and three lines suffice to calibrate the camera up to an ambiguity between focal length and radial distortion. The calibration results highly depend on accurate circle estimation, which is hard to accomplish because lines tend to project into short circular arcs. To overcome this problem, we show that, given a short circular arc edge, it is possible to robustly determine a line that goes through the center of the corresponding circle. These lines, henceforth called Lines of Circle Centres (LCCs), are used in a new method that detects sets of parallel lines and estimates the calibration parameters, including the center and amount of distortion, focal length, and camera orientation with respect to the Manhattan frame. Extensive experiments in both semi-synthetic and real images show that our algorithm outperforms state-of-the-art approaches in unsupervised calibration from a single image, while providing more information.

We introduce an unsupervised semantic scene labeling approach that continuously learns and adapts semantic models discovered within a data stream. While closely related to unsupervised video segmentation, our algorithm is not designed to be an early video processing strategy that produces coherent over-segmentations, but instead, to directly learn higher-level semantic concepts. This is achieved with an ensemble-based approach, where each learner clusters data from a local window in the data stream. Overlapping local windows are processed and encoded in a graph structure to create a label mapping across windows and reconcile the labelings to reduce unsupervised learning noise. Additionally, we iteratively learn a merging threshold criteria from observed data similarities to automatically determine the number of learned labels without human provided parameters. Experiments show that our approach semantically labels video streams with a high degree of accuracy, and achieves a better balance of under and over-segmentation entropy than existing video segmentation algorithms given similar numbers of label outputs.

Collecting well-annotated image datasets to train modern machine learning algorithms is prohibitively expensive for many tasks. One appealing alternative is rendering synthetic data where ground-truth annotations are generated automatically. Unfortunately, models trained purely on rendered images fail to generalize to real images. To address this shortcoming, prior work introduced unsupervised domain adaptation algorithms that have tried to either map representations between the two domains, or learn to extract features that are domain-invariant. In this work, we approach the problem in a new light by learning in an unsupervised manner a transformation in the pixel space from one domain to the other. Our generative adversarial network (GAN)-based method adapts source-domain images to appear as if drawn from the target domain. Our approach not only produces plausible samples, but also outperforms the state-of-the-art on a number of unsupervised domain adaptation scenarios by large margins. Finally, we demonstrate that the adaptation process generalizes to object classes unseen during training.

Part-based image classification aims at representing categories by small sets of learned discriminative parts, upon which an image representation is built. Considered as a promising avenue a decade ago, this direction has been neglected since the advent of deep neural networks. In this context, this paper brings two contributions: first, this work proceeds one step further compared to recent part-based models (PBM), focusing on how to learn parts without using any labeled data. Instead of learning a set of parts per class, as generally performed in the PBM literature, the proposed approach both constructs a partition of a given set of images into visually similar groups, and subsequently learns a set of discriminative parts per group in a fully unsupervised fashion. This strategy opens the door to the use of PBM in new applications where labeled data are typically not available, such as instance-based image retrieval. Second, this paper shows that despite the recent success of end-to-end models, explicit part learning can still boost classification performance. We experimentally show that our learned parts can help building efficient image representations, which outperform state-of-the art Deep Convolutional Neural Networks (DCNN) on both classification and retrieval tasks.

Learning based methods have shown very promising results for the task of depth estimation in single images. However, most existing approaches treat depth prediction as a supervised regression problem and as a result, require vast quantities of corresponding ground truth depth data for training. Just recording quality depth data in a range of environments is a challenging problem. In this paper, we innovate beyond existing approaches, replacing the use of explicit depth data during training with easier-to-obtain binocular stereo footage. We propose a novel training objective that enables our convo- lutional neural network to learn to perform single image depth estimation, despite the absence of ground truth depth data. Ex- ploiting epipolar geometry constraints, we generate disparity images by training our network with an image reconstruction loss. We show that solving for image reconstruction alone re- sults in poor quality depth images. To overcome this problem, we propose a novel training loss that enforces consistency be- tween the disparities produced relative to both the left and right images, leading to improved performance and robustness com- pared to existing approaches. Our method produces state of the art results for monocular depth estimation on the KITTI driving dataset, even outperforming supervised methods that have been trained with ground truth depth.

We present an unsupervised representation learning approach that compactly encodes the motion dependencies in videos. Given a pair of images from a video clip, our framework learns to predict the long-term 3D motions. To reduce the complexity of the learning framework, we propose to describe the motion as a sequence of atomic 3D flows computed with RGB-D modality. We use a Recurrent Neural Network based Encoder-Decoder framework to predict these sequences of flows. We argue that in order for the decoder to reconstruct these sequences, the encoder must learn a robust video representation that captures long-term motion dependencies and spatial-temporal relations. We demonstrate the effectiveness of our learned temporal representations on activity classification across multiple modalities and datasets such as NTU RGB+D and MSR Daily Activity 3D. Our framework is generic to any input modality, i.e., RGB, depth, and RGB-D videos.

We present an unsupervised learning framework for the task of dense 3D geometry and camera motion estimation from unstructured video sequences. In common with recent work, we use an end-to-end learning approach with view synthesis as the supervisory signal. In contrast to these works, our method is completely unsupervised, requiring only a sequence of images as input. We achieve this with a network that estimates the 6-DoF camera pose parameters of the input set, along with dense depth for a reference view using single-view inference. Our loss is constructed by projecting the nearby posed views into the reference view via the depth map. Results using the KITTI dataset demonstrate the effectiveness of our approach, which performs on par with another deep learning approach that assumes ground-truth pose information at training time.

Person re-identification is an open and challenging problem in computer vision. Existing approaches have concentrated on either designing the best feature representation or learning optimal matching metrics in a static setting where the number of cameras are fixed in a network. Most approaches have neglected the dynamic and open world nature of the re-identification problem, where a new camera may be temporarily inserted into an existing system to get additional information. To address such a novel and very practical problem, we propose an unsupervised adaptation scheme for re-identification models in a dynamic camera network. First, we formulate a domain perceptive re-identification method based on geodesic flow kernel that can effectively find the best source camera (already installed) to adapt with a newly introduced target camera, without requiring a very expensive training phase. Second, we introduce a transitive inference algorithm for re-identification that can exploit the information from best source camera to improve the accuracy across other camera pairs in a network of multiple cameras. Extensive experiments on four benchmark datasets demonstrate that the proposed approach significantly outperforms the state-of-the-art unsupervised learning based alternatives whilst being extremely efficient to compute.

Row-wise exposure delay present in CMOS cameras is responsible for skew and curvature distortions known as the rolling shutter (RS) effect while imaging under camera motion. Existing RS correction methods resort to using multiple images or tailor scene-specific correction schemes. We propose a convolutional neural network (CNN) architecture that automatically learns essential scene features from a single RS image to estimate the row-wise camera motion and undo RS distortions back to the time of first-row exposure. We employ long rectangular kernels to specifically learn the effects produced by the row-wise exposure. Experiments reveal that our proposed architecture performs better than the conventional CNN employing square kernels. Our single-image correction method fares well even operating in a frame-by-frame manner against video-based methods and performs better than scene-specific correction schemes even under challenging situations.

Given a state-of-the-art deep neural network classifier, we show the existence of a universal (image-agnostic) and very small perturbation vector that causes natural images to be misclassified with high probability. We propose a systematic algorithm for computing universal perturbations, and show that state-of-the-art deep neural networks are highly vulnerable to such perturbations, albeit being quasi-imperceptible to the human eye. We further empirically analyze these universal perturbations and show, in particular, that they generalize very well across neural networks. The surprising existence of universal perturbations reveals important geometric correlations among the high-dimensional decision boundary of classifiers. It further outlines potential security breaches with the existence of single directions in the input space that adversaries can possibly exploit to break a classifier on most natural images.

3D models provide a common ground for different representations of human bodies. In turn, robust 2D estimation has proven to be a powerful tool to obtain 3D fits "in-the-wild". However, depending on the level of detail, it can be hard to impossible to acquire labeled data for training 2D estimators on large scale. We propose a hybrid approach to this problem: with an extended version of the recently introduced SMPLify method, we obtain high quality 3D body model fits for multiple human pose datasets. Human annotators solely sort good and bad fits. This procedure leads to an initial dataset, UP-3D, with rich annotations. With a comprehensive set of experiments, we show how this data can be used to train discriminative models that produce results with an unprecedented level of detail: our models predict 31 segments and 91 landmark locations on the body. Using the 91 landmark pose estimator, we present state-of-the art results for 3D human pose and shape estimation using an order of magnitude less training data and without assumptions about gender or pose in the fitting procedure. We show that UP-3D can be enhanced with these improved fits to grow in quantity and quality, which makes the system deployable on large scale. The data, code and models are available for research purposes.

Event detection in unconstrained videos is conceived as a content-based video retrieval with two modalities: textual and visual. Given a text describing a novel event, the goal is to rank related videos accordingly. This task is zero-exemplar, no video examples are given to the novel event. Related works train a bank of concept detectors on external data sources. These detectors predict confidence scores for test videos, which are ranked and retrieved accordingly. In contrast, we learn a joint space in which the visual and textual representations are embedded. The space casts a novel event as a probability of pre-defined events. Also, it learns to measure the distance between an event and its related videos. Our model is trained end-to-end on publicly available EventNet. When applied to TRECVID Multimedia Event Detection dataset, it outperforms the state-of-the-art by a considerable margin.

Understanding traffic density from large-scale web camera (webcam) videos is a challenging problem because such videos have low spatial and temporal resolution, high occlusion and large perspective. To deeply understand traffic density, we explore both optimization based and deep learning based methods. To avoid individual vehicle detection or tracking, both methods map the dense image feature into vehicle density, one based on rank constrained regression and the other based on fully convolutional networks (FCN). The regression based method learns different weights for different blocks of the image to embed road geometry and significantly reduce the error induced by camera perspective. The FCN based method jointly estimates vehicle density and vehicle count with a residual learning framework to perform end-to-end dense prediction, allowing arbitrary image resolution, and adapting to different vehicle scales and perspectives. We analyze and compare both methods, and get insights from optimization based method to improve deep model. Since existing datasets do not cover all the challenges in our work, we collected and labelled a large-scale traffic video dataset, containing 60 million frames from 212 webcams. Both methods are extensively evaluated and compared on different counting tasks and datasets. FCN based method significantly reduces the mean absolute error (MAE) from 10.99 to 5.31 on the public dataset TRANCOS compared with the state-of-the-art baseline.

Text instance as one category of self-described objects provides valuable information for understanding and describing cluttered scenes. In this paper, we explore the task of unambiguous text localization and retrieval, to accurately localize a specific targeted text instance in a cluttered image given a natural language description that refers to it. To address this issue, first a novel recurrent Dense Text Localization Network (DTLN) is proposed to sequentially decode the intermediate convolutional representations of a cluttered scene image into a set of distinct text instance detections. Our approach avoids repeated detections at multiple scales of the same text instance by recurrently memorizing previous detections, and effectively tackles crowded text instances in close proximity. Second, we propose a Context Reasoning Text Retrieval (CRTR) model, which jointly encodes text instances and their context information through a recurrent network, and ranks localized text bounding boxes by a scoring function of context compatibility. Quantitative evaluations on standard scene text localization benchmarks and a newly collected scene text retrieval dataset demonstrate the effectiveness and advantages of our models for both scene text localization and retrieval.