ECCV 2018 论文列表

Despite decades of research, offline handwriting recognition (HWR) of degraded historical documents remains a challenging problem, which if solved could greatly improve the searchability of online cultural heritage archives. HWR models are often limited by the accuracy of the preceding steps of text detection and segmentation. Motivated by this, we present a deep learning model that jointly learns text detection, segmentation, and recognition using mostly images without detection or segmentation annotations. Our Start, Follow, Read (SFR) model is composed of a Region Proposal Network to find the start position of text lines, a novel line follower network that incrementally follows and preprocesses lines of (perhaps curved) text into dewarped images suitable for recognition by a CNN-LSTM network. SFR exceeds the performance of the winner of the ICDAR2017 handwriting recognition competition, even when not using the provided competition region annotations.

Semantic keypoints provide concise abstractions for a variety of visual understanding tasks. Existing methods define semantic keypoints separately for each category with a fixed number of semantic labels in fixed indices. As a result, this keypoint representation is in-feasible when objects have a varying number of parts, e.g. chairs with varying number of legs. We propose a category-agnostic keypoint representation, which combines a multi-peak heatmap (StarMap) for all the keypoints and their corresponding features as 3D locations in the canonical viewpoint (CanViewFeature) defined for each instance. Our intuition is that the 3D locations of the keypoints in canonical object views contain rich semantic and compositional information. Using our flexible representation, we demonstrate competitive performance in keypoint detection and localization compared to category-specific state-of-the-art methods. Moreover, we show that when augmented with an additional depth channel (DepthMap) to lift the 2D keypoints to 3D, our representation can achieve state-of-the-art results in viewpoint estimation. Finally, we show that our category-agnostic keypoint representation can be generalized to novel categories.

In this paper, we study the problem of image-text matching. Inferring the latent semantic alignment between objects or other salient stuff (e.g. snow, sky, lawn) and the corresponding words in sentences allows to capture fine-grained interplay between vision and language, and makes image-text matching more interpretable. Prior work either simply aggregates the similarity of all possible pairs of regions and words without attending differentially to more and less important words or regions, or uses a multi-step attentional process to capture limited number of semantic alignments which is less interpretable. In this paper, we present Stacked Cross Attention to discover the full latent alignments using both image regions and words in a sentence as context and infer image-text similarity. Our approach achieves the state-of-the-art results on the MS-COCO and Flickr30K datasets. On Flickr30K, our approach outperforms the current best methods by 22.1% relatively in text retrieval from image query, and 18.2% relatively in image retrieval with text query (based on Recall@1). On MS-COCO, our approach improves sentence retrieval by 17.8% relatively and image retrieval by 16.6% relatively (based on Recall@1 using the 5K test set). Code has been made available at: https://github.com/kuanghuei/SCAN.

Deep neural networks have enjoyed remarkable success for various vision tasks, however it remains challenging to apply CNNs to domains lacking a regular underlying structures such as 3D point clouds. Towards this we propose a novel convolutional architecture, termed SpiderCNN, to efficiently extract geometric features from point clouds. SpiderCNN is comprised of units called SpiderConv, which extend convolutional operations from regular grids to irregular point sets that can be embedded in R^n, by parametrizing a family of convolutional filters. We design the filter as a product of a simple step function that captures local geodesic information and a Taylor polynomial that ensures the expressiveness. SpiderCNN inherits the multi-scale hierarchical architecture from classical CNNs, which allows it to extract semantic deep features. Experiments on ModelNet40 demonstrate that SpiderCNN achieves state-of-the-art accuracy 92.4% on standard benchmarks, and shows competitive performance on segmentation task.

Omnidirectional cameras offer great benefits over classical cameras wherever a wide field of view is essential, such as in virtual reality applications or in autonomous robots. Unfortunately, standard convolutional neural networks are not well suited for this scenario as the natural projection surface is a sphere which cannot be unwrapped to a plane without introducing significant distortions, particularly in the polar regions. In this work, we present SphereNet, a novel deep learning framework which encodes invariance against such distortions explicitly into convolutional neural networks. Towards this goal, SphereNet adapts the sampling locations of the convolutional filters, effectively reversing distortions, and wraps the filters around the sphere. By building on regular convolutions, SphereNet enables the transfer of existing perspective convolutional neural network models to the omnidirectional case. We demonstrate the effectiveness of our method on the tasks of image classification and object detection, exploiting two newly created semi-synthetic and real-world omnidirectional datasets.

Most multi-view 3D reconstruction algorithms, especially when shape-from-shading cues are used, assume that object appearance is predominantly diffuse. To alleviate this restriction, we introduce S2Dnet, a generative adversarial network for transferring multiple views of objects with specular reflection into diffuse ones, so that multi-view reconstruction methods can be applied more effectively. Our network extends unsupervised image-to-image translation to multi-view ``specular to diffuse" translation. To preserve object appearance across multiple views, we introduce a Multi-View Coherence loss (MVC) that evaluates the similarity and faithfulness of local patches after the view-transformation. Our MVC loss ensures that the similarity of local correspondences among multi-view images is preserved under the image-to-image translation. As a result, our network yields significantly better results than several single-view baseline techniques. In addition, we carefully design and generate a large synthetic training data set using physically-based rendering. During testing, our network takes only the raw glossy images as input, without extra information such as segmentation masks or lighting estimation. Results demonstrate that multi-view reconstruction can be significantly improved using the images filtered by our network. We also show promising performance on real world training and testing data.

State-of-the-art video restoration methods integrate optical flow estimation networks to utilize temporal information. However, these networks typically consider only a pair of consecutive frames and hence are not capable of capturing long-range temporal dependencies and fall short of establishing correspondences across several timesteps. To alleviate these problems, we propose a novel Spatio-temporal Transformer Network (STTN) which handles multiple frames at once and thereby manages to mitigate the common nuisance of occlusions in optical flow estimation. Our proposed STTN comprises a module that estimates optical flow in both space and time and a resampling layer that selectively warps target frames using the estimated flow. In our experiments, we demonstrate the efficiency of the proposed network and show state-of-the-art restoration results in video super-resolution and video deblurring.

The work in this paper is driven by the question if spatio-temporal correlations are enough for 3D convolutional neural networks (CNN)? Most of the traditional 3D networks use local spatio-temporal features. We introduce a new block that models correlations between channels of a 3D CNN with respect to temporal and spatial features. This new block can be added as a residual unit to different parts of 3D CNNs. We name our novel block 'Spatio-Temporal Channel Correlation' (STC). By embedding this block to the current state-of-the-art architectures such as ResNext and ResNet, we improve the performance by 2-3% on the Kinetics dataset. Our experiments show that adding STC blocks to current state-of-the-art architectures outperforms the state-of-the-art methods on the HMDB51, UCF101 and Kinetics datasets. The other issue in training 3D CNNs is about training them from scratch with a huge labeled dataset to get a reasonable performance. So the knowledge learned in 2D CNNs is completely ignored. Another contribution in this work is a simple and effective technique to transfer knowledge from a pre-trained 2D CNN to a randomly initialized 3D CNN for a stable weight initialization. This allows us to significantly reduce the number of training samples for 3D CNNs. Thus, by fine-tuning this network, we beat the performance of generic and recent methods in 3D CNNs, which were trained on large video datasets, e.g. Sports-1M, and fine-tuned on the target datasets, e.g. HMDB51/UCF101.

We explore a key architectural aspect of deep convolutional neural networks: the pattern of internal skip connections used to aggregate outputs of earlier layers for consumption by deeper layers. Such aggregation is critical to facilitate training of very deep networks in an end-to-end manner. This is a primary reason for the widespread adoption of residual networks, which aggregate outputs via cumulative summation. While subsequent works investigate alternative aggregation operations (eg~concatenation), we focus on an orthogonal question: which outputs to aggregate at a particular point in the network. We propose a new internal connection structure which aggregates only a sparse set of previous outputs at any given depth. Our experiments demonstrate this simple design change offers superior performance with fewer parameters and lower computational requirements. Moreover, we show that sparse aggregation allows networks to scale more robustly to 1000+ layers, thereby opening future avenues for training long-running visual processes.

360° panoramas are a rich medium, yet notoriously difficult to visualize in the 2D image plane. We explore how intelligent rotations of a spherical image may enable content-aware projection with fewer perceptible distortions. Whereas existing approaches assume the viewpoint is fixed, intuitively some viewing angles within the sphere preserve high-level objects better than others. To discover the relationship between these optimal emph{snap angles} and the spherical panorama's content, we develop a reinforcement learning approach for the cubemap projection model. Implemented as a deep recurrent neural network, our method selects a sequence of rotation actions and receives reward for avoiding cube boundaries that overlap with important foreground objects. Our results demonstrate the impact both qualitatively and quantitatively.

A critical issue in pedestrian detection is to detect small-scale objects that will introduce feeble contrast and motion blur in images and videos, which in our opinion should partially resort to deep-rooted annotation bias. Motivated by this, we propose a novel method integrated with somatic topological line localization (TLL) and temporal feature aggregation for detecting multi-scale pedestrians, which works particularly well with small-scale pedestrians that are relatively far from the camera. Moreover, a post-processing scheme based on Markov Ran- dom Field (MRF) is introduced to eliminate ambiguities in occlusion cases. Applying with these methodologies comprehensively, we achieve best detection performance on Caltech benchmark and improve performance of small-scale objects signicantly (miss rate decreases from 74.53% to 60.79%). Beyond this, we also achieve competitive performance on CityPersons dataset and show the existence of annotation bias in KITTI dataset.

While deeper convolutional networks are needed to achieve maximum accuracy in visual perception tasks, for many inputs shallower networks are sufficient. We exploit this observation by learning to skip convolutional layers on a per-input basis. We introduce SkipNet, a modified residual network, that uses a gating network to selectively skip convolutional blocks based on the activations of the previous layer. We formulate the dynamic skipping problem in the context of sequential decision making and propose a hybrid learning algorithm that combines supervised learning and reinforcement learning to address the challenges of non-differentiable skipping decisions. We show SkipNet reduces computation by 30-90% while preserving the accuracy of the original model on four benchmark datasets and outperforms the state-of-the-art dynamic networks and static compression methods. We also qualitatively evaluate the gating policy to reveal a relationship between image scale and saliency and the number of layers skipped.

We contribute the rst large-scale dataset of scene sketches, SketchyScene, with the goal of advancing research on sketch understanding at both the object and scene level. The dataset is created through a novel and carefully designed crowdsourcing pipeline, enabling users to eciently generate large quantities realistic and diverse scene sketches. SketchyScene contains more than 29,000 scene-level sketches, 7,000+ pairs of scene templates and photos, and 11,000+ object sketches. All objects in the scene sketches have ground-truth semantic and instance masks. The dataset is also highly scalable and extensible, easily allowing augmenting and/or changing scene composition. We demonstrate the potential impact of SketchyScene by training new computational models for semantic segmentation of scene sketches and showing how the new dataset enables several applications including image retrieval, sketch colorization, editing, and captioning, etc. The dataset and code can be found at https://github.com/SketchyScene/SketchyScene.

Skeleton-based action recognition has made great progress recently, but many problems still remain unsolved. For example, the representations of skeleton sequences captured by most of the previous methods lack spatial structure information and detailed temporal dynamics features. In this paper, we propose a novel model with spatial reasoning and temporal stack learning (SR-TSL) for skeleton-based action recognition, which consists of a spatial reasoning network (SRN) and a temporal stack learning network (TSLN). The SRN can capture the high-level spatial structural information within each frame by a residual graph neural network, while the TSLN can model the detailed temporal dynamics of skeleton sequences by a composition of multiple skip-clip LSTMs. During training, we propose a clip-based incremental loss to optimize the model. We perform extensive experiments on the SYSU 3D Human-Object Interaction dataset and NTU RGB+D dataset and verify the effectiveness of each network of our model. The comparison results illustrate that our approach achieves much better results than the state-of-the-art methods.

Textual information found in scene images provides high level semantic information about the image and its context and it can be leveraged for better scene understanding. In this paper we address the problem of scene text retrieval: given a text query, the system must return all images containing the queried text. The novelty of the proposed model consists in the usage of a single shot CNN architecture that predicts at the same time bounding boxes and a compact text representation of the words in them. In this way, the text based image retrieval task can be casted as a simple nearest neighbor search of the query text representation over the outputs of the CNN over the entire image database. Our experiments demonstrate that the proposed architecture outperforms previous state-of-the-art while it offers a significant increase in processing speed.

Water bodies, such as puddles and flooded areas, on and off road pose significant risks to autonomous cars. Detecting water from moving camera is a challenging task as water surface is highly refractive, and its appearance varies with viewing angle, surrounding scene, weather conditions. In this paper, we present a water puddle detection method based on a Fully Convolutional Network (FCN) with our newly proposed Reflection Attention Units (RAUs). An RAU is a deep network unit designed to embody the physics of reflection on water surface from sky and nearby scene. To verify the performance of our proposed method, we collect 11455 color stereo images with polarizers, and 985 of left images are annotated and divided into 2 datasets: On Road (ONR) dataset and Off Road (OFR) dataset. We show that FCN-8s with RAUs improves significantly precision and recall metrics as compared to FCN-8s, DeepLab V2 and Gaussian Mixture Model (GMM). We also show that focal loss function can improve the performance of FCN-8s network due to the extreme imbalance of water versus ground classification problem.

Intrinsic image decomposition---decomposing a natural image into a set of images corresponding to different physical causes---is one of the key and fundamental problems of computer vision. Previous intrinsic decomposition approaches either address the problem in a fully supervised manner, or require multiple images of the same scene as input. These approaches are less desirable in practice, as ground truth intrinsic images are extremely difficult to acquire, and requirement of multiple images pose severe limitation on applicable scenarios. In this paper, we propose to bring the best of both worlds. We present a two stream convolutional neural network framework that is capable of learning the decomposition effectively in the absence of any ground truth intrinsic images, and can be easily extended to a (semi-)supervised setup. At inference time, our model can be easily reduced to a single stream module that performs intrinsic decomposition on a single input image. We demonstrate the effectiveness of our framework through extensive experimental study on both synthetic and real-world datasets, showing superior performance over previous approaches in both single-image and multi-image settings. Notably, our approach outperforms previous state-of-the-art single image methods while using only 50% of ground truth supervision.

We propose a sparse and low-rank reflection model for specular highlight detection and removal using a single input image. This model is motivated by the observation that the specular highlight of a natural image usually has large intensity but is rather sparsely distributed while the remaining diffuse reflection can be well approximated by a linear combination of several distinct colors with a sparse and low-rank weighting matrix. We further impose the non-negativity constraint on the weighting matrix as well as the highlight component to ensure that the model is purely additive. With this reflection model, we reformulate the task of highlight removal as a constrained nuclear norm and $l_1$-norm minimization problem which can be solved effectively by the augmented Lagrange multiplier method. Experimental results show that our method performs well on both synthetic images and many real-world examples and is competitive with previous methods, especially in some challenging scenarios featuring natural illumination, hue-saturation ambiguity and strong noises.

Edge detection is among the most fundamental vision problems for its role in perceptual grouping and its wide applications. Recent advances in representation learning have led to considerable improvements in this area. Many state of the art edge detection models are learned with fully convolutional networks (FCNs). However, FCN-based edge learning tends to be vulnerable to misaligned labels due to the delicate structure of edges. While such problem was considered in evaluation benchmarks, similar issue has not been explicitly addressed in general edge learning. In this paper, we show that label misalignment can cause considerably degraded edge learning quality, and address this issue by proposing a simultaneous edge alignment and learning framework. To this end, we formulate a probabilistic model where edge alignment is treated as latent variable optimization, and is learned end-to-end during network training. Experiments show several applications of this work, including improved edge detection with state of the art performance, and automatic refinement of noisy annotations.