CVPR 2017 论文列表

We propose a new method to estimate the 6-dof trajectory of a flying object such as a quadrotor UAV within a 3D airspace monitored using multiple fixed ground cameras. It is based on a new structure from motion formulation for the 3D reconstruction of a single moving point with known motion dynamics. Our main contribution is a new bundle adjustment procedure, which in addition to optimizing the camera poses, regularizes the point trajectory using a prior based on motion dynamics (or specifically flight dynamics). Furthermore, we can infer the underlying control input sent to the UAV's autopilot that determined its flight trajectory. Our method requires neither perfect single-view tracking nor appearance matching across views. For robustness, we allow the tracker to generate multiple detections per frame in each video. The true detections and the data association across videos is estimated using robust multi-view triangulation and subsequently refined in our bundle adjustment formulation. Quantitative evaluation on simulated data and experiments on real videos from indoor and outdoor scenes shows that our technique is superior to existing methods.

This work addresses fine-grained image classification. Our work is based on the hypothesis that when dealing with subtle differences among object classes it is critical to identify and only account for a few informative image parts, as the remaining image context may not only be uninformative but may also hurt recognition. This motivates us to formulate our problem as a sequential search for informative parts over a deep feature map produced by a deep Convolutional Neural Network (CNN). A state of this search is a set of proposal bounding boxes in the image, whose "informativeness" is evaluated by the heuristic function (H), and used for generating new candidate states by the successor function (S). The two functions are unified via a Long Short-Term Memory network (LSTM) into a new deep recurrent architecture, called HSnet. Thus, HSnet (i) generates proposals of informative image parts and (ii) fuses all proposals toward final fine-grained recognition. We specify both supervised and weakly supervised training of HSnet depending on the availability of object part annotations. Evaluation on the benchmark Caltech-UCSD Birds 200-2011 and Cars-196 datasets demonstrate our competitive performance relative to the state of the art.

In this paper, we revisit the LASSO sparse representation problem, which has been studied and used in a variety of different areas, ranging from signal processing and information theory to computer vision and machine learning. In the vision community, it found its way into many important applications, including face recognition, tracking, super resolution, image denoising, to name a few. Despite advances in efficient sparse algorithms, solving large-scale LASSO problems remains a challenge. To circumvent this difficulty, people tend to downsample and subsample the problem (e.g. via dimensionality reduction) to maintain a manageable sized LASSO, which usually comes at the cost of losing solution accuracy. This paper proposes a novel circulant reformulation of the LASSO that lifts the problem to a higher dimension, where ADMM can be efficiently applied to its dual form. Because of this lifting, all optimization variables are updated using only basic element-wise operations, the most computationally expensive of which is a 1D FFT. In this way, there is no need for a linear system solver nor matrix-vector multiplication. Since all operations in our FFTLasso method are element-wise, the subproblems are completely independent and can be trivially parallelized (e.g. on a GPU). The attractive computational properties of FFTLasso are verified by extensive experiments on synthetic and real data and on the face recognition task. They demonstrate that FFTLasso scales much more effectively than a state-of-the-art solver.

Improvements in color constancy have arisen from the use of convolutional neural networks (CNNs). However, the patch-based CNNs that exist for this problem are faced with the issue of estimation ambiguity, where a patch may contain insufficient information to establish a unique or even a limited possible range of illumination colors. Image patches with estimation ambiguity not only appear with great frequency in photographs, but also significantly degrade the quality of network training and inference. To overcome this problem, we present a fully convolutional network architecture in which patches throughout an image can carry different confidence weights according to the value they provide for color constancy estimation. These confidence weights are learned and applied within a novel pooling layer where the local estimates are merged into a global solution. With this formulation, the network is able to determine "what to learn" and "how to pool" automatically from color constancy datasets without additional supervision. The proposed network also allows for end-to-end training, and achieves higher efficiency and accuracy. On standard benchmarks, our network outperforms the previous state-of-the-art while achieving 120x greater efficiency.

Robust perception-action models should be learned from training data with diverse visual appearances and realistic behaviors, yet current approaches to deep visuomotor policy learning have been generally limited to in-situ models learned from a single vehicle or simulation environment. We advocate learning a generic vehicle motion model from large scale crowd-sourced video data, and develop an end-to-end trainable architecture for learning to predict a distribution over future vehicle egomotion from instantaneous monocular camera observations and previous vehicle state. Our model incorporates a novel FCN-LSTM architecture, which can be learned from large-scale crowd-sourced vehicle action data, and leverages available scene segmentation side tasks to improve performance under a privileged learning paradigm. We provide a novel large-scale dataset of crowd-sourced driving behavior suitable for training our model, and report results predicting the driver action on held out sequences across diverse conditions.

While the ready availability of 3D scan data has influenced research throughout computer vision, less attention has focused on 4D data; that is 3D scans of moving non-rigid objects, captured over time. To be useful for vision research, such 4D scans need to be registered, or aligned, to a common topology. Consequently, extending mesh registration methods to 4D is important. Unfortunately, no ground-truth datasets are available for quantitative evaluation and comparison of 4D registration methods. To address this we create a novel dataset of high-resolution 4D scans of human subjects in motion, captured at 60 fps. We propose a new mesh registration method that uses both 3D geometry and texture information to register all scans in a sequence to a common reference topology. The approach exploits consistency in texture over both short and long time intervals and deals with temporal offsets between shape and texture capture. We show how using geometry alone results in significant errors in alignment when the motions are fast and non-rigid. We evaluate the accuracy of our registration and provide a dataset of 40,000 raw and aligned meshes. Dynamic FAUST extends the popular FAUST dataset to dynamic 4D data, and is available for research purposes at http://dfaust.is.tue.mpg.de.

The large pose discrepancy between two face images is one of the key challenges in face recognition. Conventional approaches for pose-invariant face recognition either perform face frontalization on, or learn a pose-invariant representation from, a non-frontal face image. We argue that it is more desirable to perform both tasks jointly to allow them to leverage each other. To this end, this paper proposes Disentangled Representation learning-Generative Adversarial Network (DR-GAN) with three distinct novelties. First, the encoder-decoder structure of the generator allows DR-GAN to learn a generative and discriminative representation, in addition to image synthesis. Second, this representation is explicitly disentangled from other face variations such as pose, through the pose code provided to the decoder and pose estimation in the discriminator. Third, DR-GAN can take one or multiple images as the input, and generate one unified representation along with an arbitrary number of synthetic images. Quantitative and qualitative evaluation on both controlled and in-the-wild databases demonstrate the superiority of DR-GAN over the state of the art.

Relationships among objects play a crucial role in image understanding. Despite the great success of deep learning techniques in recognizing individual objects, reasoning about the relationships among objects remains a challenging task. Previous methods often treat this as a classification problem, considering each type of relationship (e.g. "ride") or each distinct visual phrase (e.g. "person- ride-horse") as a category. Such approaches are faced with significant difficulties caused by the high diversity of visual appearance for each kind of relationships or the large number of distinct visual phrases. We propose an integrated framework to tackle this problem. At the heart of this framework is the Deep Relational Network, a novel formulation designed specifically for exploiting the statistical dependencies between objects and their relationships. On two large data sets, the proposed method achieves substantial improvement over state-of-the-art.

Today's person detection methods work best when people are in common upright poses and appear reasonably well spaced out in the image. However, in many real images, that's not what people do. People often appear quite close to each other, e.g., with limbs linked or heads touching, and their poses are often not pedestrian-like. We propose an approach to detangle people in multi-person images. We formulate the task as a region assembly problem. Starting from a large set of overlapping regions from body part semantic segmentation and generic object proposals, our optimization approach reassembles those pieces together into multiple person instances. Since optimal region assembly is a challenging combinatorial problem, we present a Lagrangian relaxation method to accelerate the lower bound estimation, thereby enabling a fast branch and bound solution for the global optimum. As output, our method produces a pixel-level map indicating both 1) the body part labels (arm, leg, torso, and head), and 2) which parts belong to which individual person. Our results on challenging datasets show our method is robust to clutter, occlusion, and complex poses. It outperforms a variety of competing methods, including existing detector CRF methods and region CNN approaches. In addition, we demonstrate its impact on a proxemics recognition task, which demands a precise representation of "whose body part is where" in crowded images.

Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections--one between each layer and its subsequent layer--our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR- 10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less memory and computation to achieve high performance. Code and pre-trained models are available at https://github.com/liuzhuang13/DenseNet.

Image captioning is a challenging problem owing to the complexity in understanding the image content and diverse ways of describing it in natural language. Recent advances in deep neural networks have substantially improved the performance of this task. Most state-of-the-art approaches follow an encoder-decoder framework, which generates captions using a sequential recurrent prediction model. However, in this paper, we introduce a novel decision-making framework for image captioning. We utilize a "policy network" and a "value network" to collaboratively generate captions. The policy network serves as a local guidance by providing the confidence of predicting the next word according to the current state. Additionally, the value network serves as a global and lookahead guidance by evaluating all possible extensions of the current state. In essence, it adjusts the goal of predicting the correct words towards the goal of generating captions similar to the ground truth captions. We train both networks using an actor-critic reinforcement learning model, with a novel reward defined by visual-semantic embedding. Extensive experiments and analyses on the Microsoft COCO dataset show that the proposed framework outperforms state-of-the-art approaches across different evaluation metrics.

We present a CNN-based technique to estimate high-dynamic range outdoor illumination from a single low dynamic range image. To train the CNN, we leverage a large dataset of outdoor panoramas. We fit a low-dimensional physically-based outdoor illumination model to the skies in these panoramas giving us a compact set of parameters (including sun position, atmospheric conditions, and camera parameters). We extract limited field-of-view images from the panoramas, and train a CNN with this large set of input image--output lighting parameter pairs. Given a test image, this network can be used to infer illumination parameters that can, in turn, be used to reconstruct an outdoor illumination environment map. We demonstrate that our approach allows the recovery of plausible illumination conditions and enables photorealistic virtual object insertion from a single image. An extensive evaluation on both the panorama dataset and captured HDR environment maps shows that our technique significantly outperforms previous solutions to this problem.

What if we could effectively read the mind and transfer human visual capabilities to computer vision methods? In this paper, we aim at addressing this question by developing the first visual object classifier driven by human brain signals. In particular, we employ EEG data evoked by visual object stimuli combined with Recurrent Neural Networks (RNN) to learn a discriminative brain activity manifold of visual categories in a reading the mind effort. Afterward, we transfer the learned capabilities to machines by training a Convolutional Neural Network (CNN)-based regressor to project images onto the learned manifold, thus allowing machines to employ human brain-based features for automated visual classification. We use a 128-channel EEG with active electrodes to record brain activity of several subjects while looking at images of 40 ImageNet object classes. The proposed RNN-based approach for discriminating object classes using brain signals reaches an average accuracy of about 83%, which greatly outperforms existing methods attempting to learn EEG visual object representations. As for automated object categorization, our human brain-driven approach obtains competitive performance, comparable to those achieved by powerful CNN models and it is also able to generalize over different visual datasets.

Image matting is a fundamental computer vision problem and has many applications. Previous algorithms have poor performance when an image has similar foreground and background colors or complicated textures. The main reasons are prior methods 1) only use low-level features and 2) lack high-level context. In this paper, we propose a novel deep learning based algorithm that can tackle both these problems. Our deep model has two parts. The first part is a deep convolutional encoder-decoder network that takes an image and the corresponding trimap as inputs and predict the alpha matte of the image. The second part is a small convolutional network that refines the alpha matte predictions of the first network to have more accurate alpha values and sharper edges. In addition, we also create a large-scale image matting dataset including 49300 training images and 1000 testing images. We evaluate our algorithm on the image matting benchmark, our testing set, and a wide variety of real images. Experimental results clearly demonstrate the superiority of our algorithm over previous methods.

We introduce a new problem of gaze anticipation on egocentric videos. This substantially extends the conventional gaze prediction problem to future frames by no longer confining it on the current frame. To solve this problem, we propose a new generative adversarial neural network based model, Deep Future Gaze (DFG). DFG generates multiple future frames conditioned on the single current frame and anticipates corresponding future gazes in next few seconds. It consists of two networks: generator and discriminator. The generator uses a two-stream spatial temporal convolution architecture (3D-CNN) explicitly untangling the foreground and the background to generate future frames. It then attaches another 3D-CNN for gaze anticipation based on these synthetic frames. The discriminator plays against the generator by differentiating the synthetic frames of the generator from the real frames. Through competition with discriminator, the generator progressively improves quality of the future frames and thus anticipates future gaze better. Experimental results on the publicly available egocentric datasets show that DFG significantly outperforms all well-established baselines. Moreover, we demonstrate that DFG achieves better performance of gaze prediction on current frames than state-of-the-art methods. This is due to benefiting from learning motion discriminative representations in frame generation. We further contribute a new egocentric dataset (OST) in the object search task. DFG also achieves the best performance for this challenging dataset.

Watching a 360* sports video requires a viewer to continuously select a viewing angle, either through a sequence of mouse clicks or head movements. To relieve the viewer from this "360 piloting" task, we propose "deep 360 pilot" - a deep learning-based agent for piloting through 360* sports videos automatically. At each frame, the agent observes a panoramic image and has the knowledge of previously selected viewing angles. The task of the agent is to shift the current viewing angle (i.e. action) to the next preferred one (i.e., goal). We propose to directly learn an online policy of the agent from data. Specifically, we leverage a state-of-the-art object detector to propose a few candidate objects of interest (yellow boxes in Fig. 1). Then, a recurrent neural network is used to select the main object (green dash boxes in Fig. 1). Given the main object and previously selected viewing angles, our method regresses a shift in viewing angle to move to the next one. We use the policy gradient technique to jointly train our pipeline, by minimizing: (1) a regression loss measuring the distance between the selected and ground truth viewing angles, (2) a smoothness loss encouraging smooth transition in viewing angle, and (3) maximizing an expected reward of focusing on a foreground object. To evaluate our method, we built a new 360-Sports video dataset consisting of five sports domains. We trained domain-specific agents and achieved the best performance on viewing angle selection accuracy and users' preference compared to [54] and other baselines.

RANSAC is an important algorithm in robust optimization and a central building block for many computer vision applications. In recent years, traditionally hand-crafted pipelines have been replaced by deep learning pipelines, which can be trained in an end-to-end fashion. However, RANSAC has so far not been used as part of such deep learning pipelines, because its hypothesis selection procedure is non-differentiable. In this work, we present two different ways to overcome this limitation. The most promising approach is inspired by reinforcement learning, namely to replace the deterministic hypothesis selection by a probabilistic selection for which we can derive the expected loss w.r.t. to all learnable parameters. We call this approach DSAC, the differentiable counterpart of RANSAC. We apply DSAC to the problem of camera localization, where deep learning has so far failed to improve on traditional approaches. We demonstrate that by directly minimizing the expected loss of the output camera poses, robustly estimated by RANSAC, we achieve an increase in accuracy. In the future, any deep learning pipeline can use DSAC as a robust optimization component.

Correlation filter (CF) based trackers have recently gained a lot of popularity due to their impressive performance on benchmark datasets, while maintaining high frame rates. A significant amount of recent research focuses on the incorporation of stronger features for a richer representation of the tracking target. However, this only helps to discriminate the target from background within a small neighborhood. In this paper, we present a framework that allows the explicit incorporation of global context within CF trackers. We reformulate the original optimization problem and provide a closed form solution for single and multi-dimensional features in the primal and dual domain. Extensive experiments demonstrate that this framework significantly improves the performance of many CF trackers with only a modest impact on frame rate.

For machines to interact with the physical world, they must understand the physical properties of objects and materials they encounter. We use fabrics as an example of a deformable material with a rich set of mechanical properties. A thin flexible fabric, when draped, tends to look different from a heavy stiff fabric. It also feels different when touched. Using a collection of 118 fabric samples, we captured color and depth images of draped fabrics along with tactile data from a high-resolution touch sensor. We then sought to associate the information from vision and touch by jointly training CNN's across the three modalities. Through the CNN, each input, regardless of the modality, generates an embedding vector that records the fabric's physical property. By comparing the embedding vectors, our system is able to look at a fabric image and predict how it will feel, and vice versa. We also show that a system jointly trained on vision and touch data can outperform a similar system trained only on visual data when tested purely with visual inputs.