ICCV 2017 论文列表

In this paper, we propose a discriminative aggregation network (DAN) for video face recognition, which aims to integrate information from video frames effectively and efficiently. Different from existing aggregation methods, our method aggregates raw video frames directly instead of the features obtained by complex processing. By combining the idea of metric learning and adversarial learning, we learn an aggregation network that produces more discriminative synthesized images compared to input frames. Our framework reduces the number of frames to be processed and greatly speed up the recognition procedure. Furthermore, low-quality frames containing misleading information are denoised during the aggregation process, making the system more robust and discriminative. Experimental results show that our framework can generate discriminative images from video clips and improve the overall recognition performance in both the speed and accuracy on three widely used datasets.

We present a minimalistic but effective neural network that computes dense facial correspondences in highly unconstrained RGB images. Our network learns a per-pixel flow and a matchability mask between 2D input photographs of a person and the projection of a textured 3D face model. To train such a network, we generate a massive dataset of synthetic faces with dense labels using renderings of a morphable face model with variations in pose, expressions, lighting, and occlusions. We found that a training refinement using real photographs is required to drastically improve the ability to handle real images. When combined with a facial detection and 3D face fitting step, we show that our approach outperforms the state-of-the-art face alignment methods in terms of accuracy and speed. By directly estimating dense correspondences, we do not rely on the full visibility of sparse facial landmarks and are not limited to the model space of regression-based approaches. We also assess our method on video frames and demonstrate successful per-frame processing under extreme pose variations, occlusions, and lighting conditions. Compared to existing 3D facial tracking techniques, our fitting does not rely on previous frames or frontal facial initialization and is robust to imperfect face detections.

Vehicle re-identification is an important problem and has many applications in video surveillance and intelligent transportation. It gains increasing attention because of the recent advances of person re-identification techniques. However, unlike person re-identification, the visual differences between pairs of vehicle images are usually subtle and even challenging for humans to distinguish. Incorporating additional spatio-temporal information is vital for solving the challenging re-identification task. Existing vehicle re-identification methods ignored or used over-simplified models for the spatio-temporal relations between vehicle images. In this paper, we propose a two-stage framework that incorporates complex spatio-temporal information for effectively regularizing the re-identification results. Given a pair of vehicle images with their spatio-temporal information, a candidate visual-spatio-temporal path is first generated by a chain MRF model with a deeply learned potential function, where each visual-spatio-temporal state corresponds to an actual vehicle image with its spatio-temporal information. A Siamese-CNN+Path-LSTM model takes the candidate path as well as the pairwise queries to generate their similarity score. Extensive experiments and analysis show the effectiveness of our proposed method and individual components.

We introduce the first goal-driven training for visual question answering and dialog agents. Specifically, we pose a cooperative `image guessing' game between two agents -- Qbot and Abot -- who communicate in natural language dialog so that Qbot can select an unseen image from a lineup of images. We use deep reinforcement learning (RL) to end-to-end learn the policies of these agents -- from pixels to multi-agent multi-round dialog to game reward. We demonstrate two experimental results. First, as a `sanity check' demonstration of pure RL (from scratch), we show results on a synthetic world, where the agents communicate in ungrounded vocabulary, ie, symbols with no pre-specified meanings (X, Y, Z). We find that two bots invent their own communication protocol and start using certain symbols to ask/answer about certain visual attributes (shape/color/size). Thus, we demonstrate the emergence of grounded language and communication among `visual' dialog agents with no human supervision at all. Second, we conduct large-scale real-image experiments on the VisDial dataset, where we pretrain on dialog data and show that the RL fine-tuned agents significantly outperform supervised pretraining. Interestingly, the RL Qbot learns to ask questions that Abot is good at, ultimately resulting in more informative dialog and a better team.

We present an approach to learning features that represent the local geometry around a point in an unstructured point cloud. Such features play a central role in geometric registration, which supports diverse applications in robotics and 3D vision. Current state-of-the-art local features for unstructured point clouds have been manually crafted and none combines the desirable properties of precision, compactness, and robustness. We show that features with these properties can be learned from data, by optimizing deep networks that map high-dimensional histograms into low-dimensional Euclidean spaces. The presented approach yields a family of features, parameterized by dimension, that are both more compact and more accurate than existing descriptors.

As handheld video cameras are now commonplace and available in every smartphone images and videos can be recorded almost everywhere at any time. However, taking a quick shot frequently ends up in a blurry result due to unwanted camera shake during recording or moving objects in the scene. Removing these artifacts from the blurry recordings is a highly ill-posed problem as neither the sharp image nor the motion blur is known. Propagating information between multiple consecutive blurry observations can help to restore the desired sharp image or video. Solutions for blind deconvolution based on neural networks rely on a massive amount of ground-truth data which was difficult to acquire. In this work, we propose an efficient approach to produce a significant amount of realistic training data and introduce a novel recurrent network architecture to deblur frames, which can efficiently handle arbitrary spatial and temporal input sizes.

Convolutional Neural Networks (CNNs) are well established models capable of achieving state-of-the-art classification accuracy for various computer vision tasks. However, they are becoming increasingly larger, using millions of parameters, while they are restricted to handling images of fixed size. In this paper, a quantization-based approach, inspired from the well-known Bag-of-Features model, is proposed to overcome these limitations. The proposed approach, called Convolutional BoF (CBoF), uses RBF neurons to quantize the information extracted from the convolutional layers and it is able to natively classify images of various sizes as well as to significantly reduce the number of parameters in the network. In contrast to other global pooling operators and CNN compression techniques the proposed method utilizes a trainable pooling layer that it is end-to-end differentiable, allowing the network to be trained using regular back-propagation and to achieve greater distribution shift invariance than competitive methods. The ability of the proposed method to reduce the parameters of the network and increase the classification accuracy over other state-of-the-art techniques is demonstrated using three image datasets.

Correlation Filters (CFs) have recently demonstrated excellent performance in terms of rapidly tracking objects under challenging photometric and geometric variations. The strength of the approach comes from its ability to efficiently learn - "on the fly" - how the object is changing over time. A fundamental drawback to CFs, however, is that the background of the target is not modeled over time which can result in suboptimal performance. Recent tracking algorithms have suggested to resolve this drawback by either learning CFs from more discriminative deep features (e.g. DeepSRDCF and CCOT) or learning complex deep trackers (e.g. MDNet and FCNT). While such methods have been shown to work well, they suffer from high complexity: extracting deep features or applying deep tracking frameworks is very computationally expensive. This limits the real-time performance of such methods, even on high-end GPUs. This work proposes a Background-Aware CF based on hand-crafted features (HOG) that can efficiently model how both the foreground and background of the object varies over time. Our approach, like conventional CFs, is extremely computationally efficient- and extensive experiments over multiple tracking benchmarks demonstrate the superior accuracy and real-time performance of our method compared to the state-of-the-art trackers.

Depth sensors open up possibilities of dealing with the human action recognition problem by providing 3D human skeleton data and depth images of the scene. Analysis of human actions based on 3D skeleton data has become popular recently, due to its robustness and view-invariant representation. However, the skeleton alone is insufficient to distinguish actions which involve human-object interactions. In this paper, we propose a deep model which efficiently models human-object interactions and intra-class variations under viewpoint changes. First, a human body-part model is introduced to transfer the depth appearances of body-parts to a shared view-invariant space. Second, an end-to-end learning framework is proposed which is able to effectively combine the view-invariant body-part representation from skeletal and depth images, and learn the relations between the human body-parts and the environmental objects, the interactions between different human body-parts, and the temporal structure of human actions. We have evaluated the performance of our proposed model against 15 existing techniques on two large benchmark human action recognition datasets including NTU RGB+D and UWA3DII. The Experimental results show that our technique provides a significant improvement over state-of-the-art methods.

Traditional approaches for learning 3D object categories use either synthetic data or manual supervision. In this paper, we propose a method which does not require manual annotations and is instead cued by observing objects from a moving vantage point. Our system builds on two innovations: a Siamese viewpoint factorization network that robustly aligns different videos together without explicitly comparing 3D shapes; and a 3D shape completion network that can extract the full shape of an object from partial observations. We also demonstrate the benefits of configuring networks to perform probabilistic predictions as well as of geometry-aware data augmentation schemes. We obtain state-of-the-art results on publicly-available benchmarks.

Learned boundary maps are known to outperform hand-crafted ones as a basis for the watershed algorithm. We show, for the first time, how to train watershed computation jointly with boundary map prediction. The estimator for the merging priorities is cast as a neural network that is convolutional (over space) and recurrent (over iterations). The latter allows learning of complex shape priors. The method gives the best known seeded segmentation results on the CREMI segmentation challenge.

Estimating a depth map from multiple views of a scene is a fundamental task in computer vision. As soon as more than two viewpoints are available, one faces the very basic question how to measure similarity across >2 image patches. Surprisingly, no direct solution exists, instead it is common to fall back to more or less robust averaging of two-view similarities. Encouraged by the success of machine learning, and in particular convolutional neural networks, we propose to learn a matching function which directly maps multiple image patches to a scalar similarity score. Experiments on several multi-view datasets demonstrate that this approach has advantages over methods based on pairwise patch similarity.

Human actions captured in video sequences are three-dimensional signals characterizing visual appearance and motion dynamics. To learn action patterns, existing methods adopt Convolutional and/or Recurrent Neural Networks (CNNs and RNNs). CNN based methods are effective in learning spatial appearances, but are limited in modeling long-term motion dynamics. RNNs, especially Long Short-Term Memory (LSTM), are able to learn temporal motion dynamics. However, naively applying RNNs to video sequences in a convolutional manner implicitly assumes that motions in videos are stationary across different spatial locations. This assumption is valid for short-term motions but invalid when the duration of the motion is long. In this work, we propose Lattice-LSTM, which extends LSTM by learning independent hidden state transitions of memory cells for individual spatial locations. This method effectively enhances the ability to model dynamics across time and addresses the non-stationary issue of long-term motion dynamics without significantly increasing the model complexity. Additionally, we introduce a novel multi-modal training procedure for training our network. Unlike traditional two-stream architectures which use RGB and optical flow information as input, our two-stream model leverages both modalities to jointly train both input gates and both forget gates in the network rather than treating the two streams as separate entities with no information about the other. We apply this end-to-end system to benchmark datasets (UCF-101 and HMDB-51) of human action recognition. Experiments show that on both datasets, our proposed method outperforms all existing ones that are based on LSTM and/or CNNs of similar model complexities.

We propose an attentive local feature descriptor suitable for large-scale image retrieval, referred to as DELF (DEep Local Feature). The new feature is based on convolutional neural networks, which are trained only with image-level annotations on a landmark image dataset. To identify semantically useful local features for image retrieval, we also propose an attention mechanism for keypoint selection, which shares most network layers with the descriptor. This framework can be used for image retrieval as a drop-in replacement for other keypoint detectors and descriptors, enabling more accurate feature matching and geometric verification. Our system produces reliable confidence scores to reject false positives---in particular, it is robust against queries that have no correct match in the database. To evaluate the proposed descriptor, we introduce a new large-scale dataset, referred to as Google-Landmarks dataset, which involves challenges in both database and query such as background clutter, partial occlusion, multiple landmarks, objects in variable scales, etc.

3D face reconstruction is a fundamental Computer Vision problem of extraordinary difficulty. Current systems often assume the availability of multiple facial images (sometimes from the same subject) as input, and must address a number of methodological challenges such as establishing dense correspondences across large facial poses, expressions, and non-uniform illumination. In general these methods require complex and inefficient pipelines for model building and fitting. In this work, we propose to address many of these limitations by training a Convolutional Neural Network (CNN) on an appropriate dataset consisting of 2D images and 3D facial models or scans. Our CNN works with just a single 2D facial image, does not require accurate alignment nor establishes dense correspondence between images, works for arbitrary facial poses and expressions, and can be used to reconstruct the whole 3D facial geometry (including the non-visible parts of the face) bypassing the construction (during training) and fitting (during testing) of a 3D Morphable Model. We achieve this via a simple CNN architecture that performs direct regression of a volumetric representation of the 3D facial geometry from a single 2D image. We also demonstrate how the related task of facial landmark localization can be incorporated into the proposed framework and help improve reconstruction quality, especially for the cases of large poses and facial expressions.

This paper addresses the problem of jointly recognizing object fluents and tasks in egocentric videos. Fluents are the changeable attributes of objects. Tasks are goal-oriented human activities which interact with objects and aim to change some attributes of the objects. The process of executing a task is a process to change the object fluents over time. We propose a hierarchical model to represent tasks as concurrent and sequential object fluents. In a task, different fluents closely interact with each other both in spatial and temporal domains. Given an egocentric video, a beam search algorithm is applied to jointly recognizing the object fluents in each frame, and the task of the entire video. We collected a large scale egocentric video dataset of tasks and fluents. This dataset contains 14 categories of tasks, 25 object classes, 21 categories of object fluents, 809 video sequences, and approximately 333,000 video frames. The experimental results on this dataset prove the strength of our method.

Person Re-Identification (person re-id) is a crucial task as its applications in visual surveillance and human-computer interaction. In this work, we present a novel joint Spatial and Temporal Attention Pooling Network (ASTPN) for video-based person re-identification, which enables the feature extractor to be aware of the current input video sequences, in a way that interdependency from the matching items can directly influence the computation of each other's representation. Specifically, the spatial pooling layer is able to select regions from each frame, while the attention temporal pooling performed can select informative frames over the sequence, both pooling guided by the information from distance matching. Experiments are conduced on the iLIDS-VID, PRID-2011 and MARS datasets and the results demonstrate that this approach outperforms existing state-of-art methods. We also analyze how the joint pooling in both dimensions can boost the person re-id performance more effectively than using either of them separately.

Most of the existing works on human activity analysis focus on recognition or early recognition of the activity labels from complete or partial observations. Predicting the labels of future unobserved activities where no frames of the predicted activities have been observed is a challenging problem, with important applications, which has not been explored much. Associated with the future label prediction problem is the problem of predicting the starting time of the next activity. In this work, we propose a system that is able to infer about the labels and the starting times of future activities. Activities are characterized by the previous activity sequence (which is observed), as well as the objects present in the scene during their occurrence. We propose a network similar to a hybrid Siamese network with three branches to jointly learn both the future label and the starting time. The first branch takes visual features from the objects present in the scene using a fully connected network, the second branch takes previous activity features using a LSTM network to model long-term sequential relationships and the third branch captures the last observed activity features to model the context of inter-activity time using another fully connected network. These concatenated features are used for both label and time prediction. Experiments on two challenging datasets demonstrate that our framework for joint prediction of activity label and starting time improves the performance of both, and outperforms the state-of-the-arts.

While most existing approaches for detection in videos focus on objects or human actions separately, we aim at jointly detecting objects performing actions, such as cat eating or dog jumping. We introduce an end-to-end multitask objective that jointly learns object-action relationships. We compare it with different training objectives, validate its effectiveness for detecting objects-actions in videos, and show that both tasks of object and action detection benefit from this joint learning. Moreover, the proposed architecture can be used for zero-shot learning of actions: our multitask objective leverages the commonalities of an action performed by different objects, eg. dog and cat jumping, enabling to detect actions of an object without training with these object-actions pairs. In experiments on the A2D dataset, we obtain state-of-the-art results on segmentation of object-action pairs. We finally apply our multitask architecture to detect visual relationships between objects in images of the VRD dataset.