ICCV 2017 论文列表

We present a new point set registration method with global-local correspondence and transformation estimation (GL-CATE). The geometric structures of point sets are exploited by combining the global feature, the point-to-point Euclidean distance, with the local feature, the shape distance (SD) which is based on the histograms generated by an elliptical Gaussian soft count strategy. By using a bi-directional deterministic annealing scheme to directly control the searching ranges of the two features, the mixture-feature Gaussian mixture model (MGMM) is constructed to recover the correspondences of point sets. A new vector based structure constraint term is formulated to regularize the transformation. The accuracy of transformation updating is improved by constraining spatial structure at both global and local scales. An annealing scheme is applied to progressively decrease the strength of the regularization and to achieve the maximum overlap. Both of the aforementioned processes are incorporated in the EM algorithm, an unified optimization framework. We test the performances of our GL-CATE in contour registration, sequence images, real images, medical images, fingerprint images and remote sensing images, and compare with eight state-of-the-art methods where our method shows favorable performances in most scenarios.

We propose a novel video object segmentation algorithm based on pixel-level matching using Convolutional Neural Networks (CNN). Our network aims to distinguish the target area from the background on the basis of the pixel-level similarity between two object units. The proposed network represents a target object using features from different depth layers in order to take advantage of both the spatial details and the category-level semantic information. Furthermore, we propose a feature compression technique that drastically reduces the memory requirements while maintaining the capability of feature representation. Two-stage training (pre-training and fine-tuning) allows our network to handle any target object regardless of its category (even if the object's type does not belong to the pre-training data) or of variations in its appearance through a video sequence. Experiments on large datasets demonstrate the effectiveness of our model - against related methods - in terms of accuracy, speed, and stability. Finally, we introduce the transferability of our network to different domains, such as the infrared data domain.

Super resolution is the problem of artificially enlarging a low resolution photograph to recover a plausible high resolution version. In the regime of high magnification factors, the problem is dramatically underspecified and many plausible, high resolution images may match a given low resolution image. In particular, traditional super resolution techniques fail in this regime due to the multimodality of the problem and strong prior information that must be imposed on image synthesis to produce plausible high resolution images. In this work we propose a new probabilistic deep network architecture, a pixel recursive super resolution model, that is an extension of PixelCNNs to address this problem. We demonstrate that this model produces a diversity of plausible high resolution images at large magnification factors. Furthermore, in human evaluation studies we demonstrate how previous methods fail to fool human observers. However, high resolution images sampled from this probabilistic deep network do fool a naive human observer a significant fraction of the time.

This paper presents a framework for localization or grounding of phrases in images using a large collection of linguistic and visual cues. We model the appearance, size, and position of entity bounding boxes, adjectives that contain attribute information, and spatial relationships between pairs of entities connected by verbs or prepositions. Special attention is given to relationships between people and clothing or body part mentions, as they are useful for distinguishing individuals. We automatically learn weights for combining these cues and at test time, perform joint inference over all phrases in a caption. The resulting system produces state of the art performance on phrase localization on the Flickr30k Entities dataset and visual relationship detection on the Stanford VRD dataset.

Automatic image aesthetics rating has received a growing interest with the recent breakthrough in deep learning. Although many studies exist for learning a generic or universal aesthetics model, investigation of aesthetics models incorporating individual user's preference is quite limited. We address this personalized aesthetics problem by showing that individual's aesthetic preferences exhibit strong correlations with content and aesthetic attributes, and hence the deviation of individual's perception from generic image aesthetics is predictable. To accommodate our study, we first collect two distinct datasets, a large image dataset from Flickr and annotated by Amazon Mechanical Turk, and a small dataset of real personal albums rated by owners. We then propose a new approach to personalized aesthetics learning that can be trained even with a small set of annotated images from a user. The approach is based on a residual-based model adaptation scheme which learns an offset to compensate for the generic aesthetics score. Finally, we introduce an active learning algorithm to optimize personalized aesthetics prediction for real-world application scenarios. Experiments demonstrate that our approach can effectively learn personalized aesthetics preferences, and outperforms existing methods on quantitative comparisons.

Cinemagraphs are a compelling way to convey dynamic aspects of a scene. In these media, dynamic and still elements are juxtaposed to create an artistic and narrative experience. Creating a high-quality, aesthetically pleasing cinemagraph requires isolating objects in a semantically meaningful way and then selecting good start times and looping periods for those objects to minimize visual artifacts (such a tearing). To achieve this, we present a new technique that uses object recognition and semantic segmentation as part of an optimization method to automatically create cinemagraphs from videos that are both visually appealing and semantically meaningful. Given a scene with multiple objects, there are many cinemagraphs one could create. Our method evaluates these multiple candidates and presents the best one, as determined by a model trained to predict human preferences in a collaborative way. We demonstrate the effectiveness of our approach with multiple results and a user study.

Input binarization has shown to be an effective way for network acceleration. However, previous binarization scheme could be regarded as simple pixel-wise thresholding operations (i.e., order-one approximation) and suffers a big accuracy loss. In this paper, we propose a high-order binarization scheme, which achieves more accurate approximation while still possesses the advantage of binary operation. In particular, the proposed scheme recursively performs residual quantization and yields a series of binary input images with decreasing magnitude scales. Accordingly, we propose high-order binary filtering and gradient propagation operations for both forward and backward computations. Theoretical analysis shows approximation error guarantee property of proposed method. Extensive experimental results demonstrate that the proposed scheme yields great recognition accuracy while being accelerated.

To bridge the gap between humans and machines in image understanding and describing, we need further insight into how people describe a perceived scene. In this paper, we study the agreement between bottom-up saliency-based visual attention and object referrals in scene description constructs. We investigate the properties of human-written descriptions and machine-generated ones. We then propose a saliency-boosted image captioning model in order to investigate benefits from low-level cues in language models. We learn that (1) humans mention more salient objects earlier than less salient ones in their descriptions, (2) the better a captioning model performs, the better attention agreement it has with human descriptions, (3) the proposed saliency-boosted model, compared to its baseline form, does not improve significantly on the MS COCO database, indicating explicit bottom-up boosting does not help when the task is well learnt and tuned on a data, (4) a better generalization is, however, observed for the saliency-boosted model on unseen data.

Progress in Multiple Object Tracking (MOT) has been limited by the size of the available datasets. We present an efficient framework to annotate trajectories and use it to produce a MOT dataset of unprecedented size. A novel path supervision paradigm lets the annotator loosely track the object with a cursor while watching the video. This results in a path annotation for each object in the sequence. These path annotations, together with object detections, are fed into a two-step optimization to produce full bounding-box trajectories. Our experiments on existing datasets prove that our framework produces more accurate annotations than the state of the art and this in a fraction of the time. We further validate our approach by generating the PathTrack dataset, with more than 15,000 person trajectories in 720 sequences. We believe tracking approaches can benefit from a larger dataset like this one, just as was the case in object recognition. We show its potential by using it to re-train an off-the-shelf person matching network, originally trained on the MOT15 dataset, almost halving the misclassification rate. Additionally, training on our data consistently improves tracking results, both on our dataset and on MOT15. In the latter, where we improve the top-performing tracker (NOMT) dropping the number of ID Switches by 18% and fragments by 5%.

At the core of many Computer Vision applications stands the need to define a mathematical model describing the imaging process. To this end, the pinhole model with radial distortion is probably the most commonly used, as it balances low complexity with a precision that is sufficient for most applications. On the other hand, unconstrained non-parametric models, despite being originally proposed to handle specialty cameras, have been shown to outperform the pinhole model, even with the simpler setups. Still, notwithstanding the higher accuracy, the inability of describing the imaging model by simple linear projective operators severely limits the use of standard algorithms with unconstrained models. In this paper we propose a parameter-free camera model where each imaging ray is constrained to a common optical center, forcing the camera to be central. Such model can be easily calibrated with a practical procedure which provides a convenient undistortion map that can be used to obtain a virtual pinhole camera. The proposed method can also be used to calibrate a stereo rig with a displacement map that simultaneously provides stereo rectification and corrects lens distortion.

Being intensively studied, visual tracking has seen great recent advances in either speed (e.g., with correlation filters) or accuracy (e.g., with deep features). Real-time and high accuracy tracking algorithms, however, remain scarce. In this paper we study the problem from a new perspective and present a novel parallel tracking and verifying (PTAV) framework, by taking advantage of the ubiquity of multi-thread techniques and borrowing from the success of parallel tracking and mapping in visual SLAM. Our PTAV framework typically consists of two components, a tracker T and a verifier V, working in parallel on two separate threads. The tracker T aims to provide a super real-time tracking inference and is expected to perform well most of the time; by contrast, the verifier V checks the tracking results and corrects T when needed. The key innovation is that, V does not work on every frame but only upon the requests from T; on the other end, T may adjust the tracking according to the feedback from V. With such collaboration, PTAV enjoys both the high efficiency provided by T and the strong discriminative power by V. In our extensive experiments on popular benchmarks including OTB2013, OTB2015, TC128 and UAV20L, PTAV achieves the best tracking accuracy among all real-time trackers, and in fact performs even better than many deep learning based solutions. Moreover, as a general framework, PTAV is very flexible and has great rooms for improvement and generalization.

We propose a deep network architecture for the pan-sharpening problem called PanNet. We incorporate domain-specific knowledge to design our PanNet architecture by focusing on the two aims of the pan-sharpening problem: spectral and spatial preservation. For spectral preservation, we add up-sampled multispectral images to the network output, which directly propagates the spectral information to the reconstructed image. To preserve spatial structure, we train our network parameters in the high-pass filtering domain rather than the image domain. We show that the trained network generalizes well to images from different satellites without needing retraining. Experiments show significant improvement over state-of-the-art methods visually and in terms of standard quality metrics.

This paper introduces a probabilistic latent variable model to address unsupervised domain adaptation problems. This is achieved by learning projections from each domain to a latent space along the classifier in the latent space to simultaneously minimizing a notion of domain disparity while maximizing a measure of discriminatory power. The non-parametric nature of our Latent variable model makes it possible to infer the latent space dimension automatically from data. We also develop a Variational Bayes (VB) algorithm for parameter estimation. We evaluate and contrast our proposed model against state-of-the-art methods for the task of visual domain adaptation using both handcrafted and deep net features. Our experiments show that even with a simple softmax classifier, our model can outperform several state-of-the-art methods taking advantage of more sophisticated classification schemes.

We aim to tackle a novel vision task called Weakly Supervised Visual Relation Detection (WSVRD) to detect "subject-predicate-object" relations in an image with object relation groundtruths available only at the image level. This is motivated by the fact that it is extremely expensive to label the combinatorial relations between objects at the instance level. Compared to the extensively studied problem, Weakly Supervised Object Detection (WSOD), WSVRD is more challenging as it needs to examine a large set of regions pairs, which is computationally prohibitive and more likely stuck in a local optimal solution such as those involving wrong spatial context. To this end, we present a Parallel, Pairwise Region-based, Fully Convolutional Network (PPR-FCN) for WSVRD. It uses a parallel FCN architecture that simultaneously performs pair selection and classification of single regions and region pairs for object and relation detection, while sharing almost all computation shared over the entire image. In particular, we propose a novel position-role-sensitive score map with pairwise RoI pooling to efficiently capture the crucial context associated with a pair of objects. We demonstrate the superiority of PPR-FCN over all baselines in solving the WSVRD challenge by using results of extensive experiments over two visual relation benchmarks.

In this paper, we tackle the vehicle Re-identification (ReID) problem which is of great importance in urban surveillance and can be used for multiple applications. In our vehicle ReID framework, an orientation invariant feature embedding module and a spatial-temporal regularization module are proposed. With orientation invariant feature embedding, local region features of different orientations can be extracted based on 20 key point locations and can be well aligned and combined. With spatial-temporal regularization, the log-normal distribution is adopted to model the spatial-temporal constraints and the retrieval results can be refined. Experiments are conducted on public vehicle ReID datasets and our proposed method achieves state-of-the-art performance. Investigations of the proposed framework is conducted, including the landmark regressor and comparisons with attention mechanism. Both the orientation invariant feature embedding and the spatio-temporal regularization achieve considerable improvements.

This paper addresses the problem of estimating the geometric transformation relating two distinct visual modalities (e.g. an image and a map, or a projective structure and a Euclidean 3D model) while relying only on semantic cues, such as semantically segmented regions or object bounding boxes. The proposed approach differs from the traditional feature-to-feature correspondence reasoning: starting from semantic regions on one side, we seek their possible corresponding regions on the other, thus constraining the sought geometric transformation. This entails a simultaneous search for the transformation and for the region-to-region correspondences.This paper is the first to derive the conditions that must be satisfied for a convex region, defined by control points, to be transformed inside an ellipsoid. These conditions are formulated as Linear Matrix Inequalities and used within a Branch-and-Prune search to obtain the globally optimal transformation. We tested our approach, under mild initial bound conditions, on two challenging registration problems for aligning: (i) a semantically segmented image and a map via a 2D homography; (ii) a projective 3D structure and its Euclidean counterpart.

Recognizing arbitrary objects in the wild has been a challenging problem due to the limitations of existing classification models and datasets. In this paper, we propose a new task that aims at parsing scenes with a large and open vocabulary, and several evaluation metrics are explored for this problem. Our approach is a joint image pixel and word concept embeddings framework, where word concepts are connected by semantic relations. We validate the open vocabulary prediction ability of our framework on ADE20K dataset which covers a wide variety of scenes and objects. We further explore the trained joint embedding space to show its interpretability.

Video object detection is a fundamental tool for many applications. Since direct application of image-based object detection cannot leverage the rich temporal information inherent in video data, we advocate to the detection of long-range video object pattern. While the Long Short-Term Memory (LSTM) has been the de facto choice for such detection, currently LSTM cannot fundamentally model object association between consecutive frames. In this paper, we propose the association LSTM to address this fundamental association problem. Association LSTM not only regresses and classifiy directly on object locations and categories but also associates features to represent each output object. By minimizing the matching error between these features, we learn how to associate objects in two consecutive frames. Additionally, our method works in an online manner, which is important for most video tasks. Compared to the traditional video object detection methods, our approach outperforms them on standard video datasets.

State-of-the-art video deblurring methods are capable of removing non-uniform blur caused by unwanted camera shake and/or object motion in dynamic scenes. However, most existing methods are based on batch processing and thus need access to all recorded frames, rendering them computationally demanding and time-consuming and thus limiting their practical use. In contrast, we propose an online (sequential) video deblurring method based on a spatio-temporal recurrent network that allows for real-time performance. In particular, we introduce a novel architecture which extends the receptive field while keeping the overall size of the network small to enable fast execution. In doing so, our network is able to remove even large blur caused by strong camera shake and/or fast moving objects. Furthermore, we propose a novel network layer that enforces temporal consistency between consecutive frames by dynamic temporal blending which compares and adaptively (at test time) shares features obtained at different time steps. We show the superiority of the proposed method in an extensive experimental evaluation.