ECCV 2018 论文列表

We address the task of jointly determining what a person is doing and where they are looking based on the analysis of video captured by a headworn camera. We propose a novel deep model for joint gaze estimation and action recognition in First Person Vision. Our method describes the participant's gaze as a probabilistic variable and models its distribution using stochastic units in a deep network. We sample from these stochastic units to generate an attention map. This attention map guides the aggregation of visual features in action recognition, thereby providing coupling between gaze and action. We evaluate our method on the standard EGTEA dataset and demonstrate performance that exceeds the state-of-the-art by a significant margin of 3.5%.

Self-supervised learning of convolutional neural networks can harness large amounts of cheap unlabeled data to train powerful feature representations. As surrogate task, we jointly address ordering of visual data in the spatial and temporal domain. The permutations of training samples, which are at the core of self-supervision by ordering, have so far been sampled randomly from a fixed preselected set. Based on deep reinforcement learning we propose a sampling policy that adapts to the state of the network, which is being trained. Therefore, new permutations are sampled according to their expected utility for updating the convolutional feature representation. Experimental evaluation on unsupervised and transfer learning tasks demonstrates competitive performance on standard benchmarks for image and video classification and nearest neighbor retrieval.

Unsupervised image-to-image translation techniques are able to map local texture between two domains, but they are typically un- successful when the domains require larger shape change. Inspired by semantic segmentation, we introduce a discriminator with dilated convo- lutions which is able to use information from across the entire image to train a more context-aware generator. This is coupled with a multi-scale perceptual loss which is better able to represent error in the underly- ing shape of objects. We demonstrate that this design is more capable of representing shape deformation in a challenging toy dataset, plus in complex mappings with significant dataset variation between humans, dolls, and anime faces, and between cats and dogs.

It is now much easier than ever before to produce videos. While the ubiquitous video data is a great source for information discovery and extraction, the computational challenges are unparalleled. Automatically summarizing the videos has become a substantial need for browsing, searching, and indexing visual content. This paper is in the vein of supervised video summarization using sequential determinantal point processes (SeqDPPs), which models diversity by a probabilistic distribution. We improve this model in two folds. In terms of learning, we propose a large-margin algorithm to address the exposure bias problem in SeqDPP. In terms of modeling, we design a new probabilistic distribution such that, when it is integrated into SeqDPP, the resulting model accepts user input about the expected length of the summary. Moreover, we also significantly extend a popular video summarization dataset by 1) more egocentric videos, 2) dense user annotations, and 3) a refined evaluation scheme. We conduct extensive experiments on this dataset (about 60 hours of videos in total) and compare our approach to several competitive baselines.

Current visual recognition is dominated by the end-to-end formulation of classification problems implemented by the parametric softmax classifiers. Such formulation makes a closed world assumption with a fixed set of categories. This becomes problematic for open-set scenarios where new categories are encountered with very few examples for learning a generalizable parametric classifier. This paper adopts a non-parametric approach for visual recognition by optimizing feature embeddings instead of parametric classifiers. We use a deep neural network to learn embeddings which preserves neighborhood structures by neighborhood component analysis (NCA). Limited by its computational bottlenecks, we devise a mechanism to use an augmented memory to scale NCA for large datasets and very deep neural networks. Our experimental results show state-of-the-art results on ImageNet classification using nearest neighbor classifiers. More importantly, our feature embedding is more generalizable for new categories such as sub-category discovery and few-shot recognition.

Person re-identification is an important task that requires learning discriminative visual features for distinguishing different person identities. Diverse auxiliary information has been utilized to improve the visual feature learning. In this paper, we propose to exploit natural language description as additional training supervisions for more effective features. Compared with other auxiliary information, language can describe a specific person from more compact and semantic visual aspects, thus is complementary to the pixel-level image data. Our method not only learns better global visual feature with the supervision of the overall description but also enforces semantic consistencies between local visual and linguistic features, which is achieved by building global and local image-language associations. The global image-language association is established according to the identity labels, while the local association is based upon the implicit correspondences between image regions and noun phrases. Extensive experiments demonstrate the effectiveness of employing language as training supervisions with the two association schemes. Our method achieves state-of-the-art performance without utilizing any auxiliary information during testing and shows better performance than other joint embedding methods for the image-language association.

Deep neural networks have lately shown tremendous performance in various applications including vision and speech processing tasks. However, alongside their ability to perform these tasks with such high accuracy, it has been shown that they are highly susceptible to adversarial attacks: a small change in the input would cause the network to err with high confidence. This phenomenon exposes an inherent fault in these networks and their ability to generalize well. For this reason, providing robustness to adversarial attacks is an important challenge in networks training, which has led to extensive research. In this work, we suggest a theoretically inspired novel approach to improve the networks' robustness. Our method applies regularization using the Frobenius norm of the Jacobian of the network, which is applied as post-processing, after regular training has finished. We demonstrate empirically that it leads to enhanced robustness results with a minimal change in the original network's accuracy.

In this paper, we present an incremental structure from motion (SfM) algorithm that significantly outperforms existing algorithms when fiducial markers are present in the scene, and that matches the performance of existing algorithms when no markers are present. Our algorithm uses markers to limit potential incorrect image matches, change the order in which images are added to the reconstruction, and enforce new bundle adjustment constraints. To validate our algorithm, we introduce a new dataset with 16 image collections of large indoor scenes with challenging characteristics (e.g., blank hallways, glass facades, brick walls) and with markers placed throughout. We show that our algorithm produces complete, accurate reconstructions on all 16 image collections, most of which cause other algorithms to fail. Further, by selectively masking fiducial markers, we show that the presence of even a small number of markers can improve the results of our algorithm.

We propose a real-time RGB-based pipeline for object detection and 6D pose estimation. Our novel 3D orientation estimation is based on a variant of the Denoising Autoencoder that is trained on simulated views of a 3D model using Domain Randomization. This so-called Augmented Autoencoder has several advantages over existing methods: It does not require real, pose-annotated training data, generalizes to various test sensors and inherently handles object and view symmetries. Instead of learning an explicit mapping from input images to object poses, it provides an implicit representation of object orientations defined by samples in a latent space. Experiments on the T-LESS and LineMOD datasets show that our method outperforms similar model-based approaches and competes with state-of-the art approaches that require real pose-annotated images.

Imagining a scene described in natural language with realistic layout and appearance of entities is the ultimate test of spatial, visual, and semantic world knowledge. As a step towards this goal, we present the Composition Retrieval and Fusion Networks (CRAFT), a model capable of learning this knowledge from video-caption data and applying it for generating videos from novel captions. CRAFT explicitly predicts a temporal-layout of mentioned entities (characters and objects), retrieves spatio-temporal entity segments from a video database and fuses them to generate scene videos. Our modeling contributions include sequential training of components of CRAFT while jointly modeling layout and appearances, and losses that encourage learning compositional representations for retrieval. We evaluate CRAFT on semantic fidelity to caption, composition consistency and visual quality. CRAFT outperforms direct pixel generation approaches and generalizes well to unseen captions as well as unseen video databases with no text annotations. We demonstrate CRAFT on FLINTSTONES, a new richly annotated video-caption dataset with over 25000 videos.

Convolutional neural network (CNN) depth is of crucial importance for image super-resolution (SR). However, we observe that deeper networks for image SR are more difficult to train. The low-resolution (LR) inputs and features contain abundant low-frequency information, which is treated equally across channels, hence hindering the representational ability of CNNs. To solve these problems, we propose the very deep residual channel attention networks (RCAN). Specifically, we propose residual in residual (RIR) structure to form very deep network, which consists of several residual groups with long skip connections. Each residual group contains some residual blocks with short skip connections. Meanwhile, RIR allows abundant low-frequency information to be bypassed through multiple skip connections, making the main network focus on learning high-frequency information. Furthermore, we propose channel attention mechanism to adaptively rescale channel-wise features by considering interdependencies among channels. Extensive experiments show that our RCAN achieves better accuracy and visual improvements against state-of-the-art methods.

This paper addresses the problem of reassembling images from disjointed fragments. More specifically, given an unordered set of fragments, we aim at reassembling one or several possibly incomplete images. The main contributions of this work are: 1) several deep neural architectures to predict the relative position of image fragments that outperform the previous state of the art; 2) casting the reassembly problem into the shortest path in a graph problem for which we provide several construction algorithms depending on available information; 3) a new dataset of images taken from the Metropolitan Museum of Art (MET) dedicated to image reassembly for which we provide a clear setup and a strong baseline.

Systems that perform image manipulation using deep convolutional networks have achieved remarkable realism. Perceptual losses and losses based on adversarial discriminators are the two main classes of learning objectives behind these advances. In this work, we show how these two ideas can be combined in a principled and non-additive manner for unaligned image translation tasks. This is accomplished through a special architecture of the discriminator network inside generative adversarial learning framework. The new architecture, that we call a perceptual discriminator, embeds the convolutional parts of a pre-trained deep classification network inside the discriminator network. The resulting architecture can be trained on unaligned image datasets, while benefiting from the robustness and efficiency of perceptual losses. We demonstrate the merits of the new architecture in a series of qualitative and quantitative comparisons with baseline approaches and state-of-the-art frameworks for unaligned image translation.

Existing deep learning based image inpainting methods use a standard convolutional network over the corrupted image, using convolutional filter responses conditioned on both valid pixels as well as the substitute values in the masked holes (typically the mean value). This often leads to artifacts such as color discrepancy and blurriness. Post-processing is usually used to reduce such artifacts, but are expensive and may fail. We propose to use partial convolutions, where the convolution is masked and renormalized to be conditioned on only valid pixels. We further include a mechanism to automatically generate an updated mask for the next layer as part of the forward pass. Our model outperforms other methods for irregular masks. We show qualitative and quantitative comparisons with other inpainting methods to validate our approach.

In this paper we investigate image generation guided by hand sketch. When the input sketch is badly drawn, the output of common image-to-image translation follows the input edges due to the hard condition imposed by the translation process. Instead, we propose to use sketch as weak constraint, where the output edges do not necessarily follow the input edges. We address this problem using a novel joint image completion approach, where the sketch provides the image context for completing, or generating the output image. We train a generated adversarial network, i.e., contextual GAN to learn the joint distribution of sketch and the corresponding image by using joint images. Our contextual GAN has several advantages. First, the simple joint image representation allows for simple and effective learning of joint distribution in the same image-sketch space, which avoids complicated issues in cross-domain learning. Second, while the output is related to its input overall, the generated features exhibit more freedom in appearance and do not strictly align with the input features as previous conditional GANs do. Third, from the joint image's point of view, image and sketch are of no difference, thus exactly the same deep joint image completion network can be used for image-to-sketch generation. Experiments evaluated on three different datasets show that our contextual GAN can generate more realistic images than state-of-the-art conditional GANs on challenging inputs and generalize well on common categories.

3D object geometry reconstruction remains a challenge when working with transparent, occluded, or highly reflective surfaces. While recent methods classify shape features using raw audio, we present a multimodal neural network optimized for estimating an object's geometry and material. Our networks use spectrograms of recorded and synthesized object impact sounds and voxelized shape estimates to extend the capabilities of vision-based reconstruction. We evaluate our method on multiple datasets of both recorded and synthesized sounds. We further present an interactive application for real-time scene reconstruction in which a user can strike objects, producing sound that can instantly classify and segment the struck object, even if the object is transparent or visually occluded.

We focus on the challenging task of real-time semantic segmentation in this paper. It finds many practical applications and yet is with fundamental difficulty of reducing a large portion of computation for pixel-wise label inference. We propose an image cascade network (ICNet) that incorporates multi-resolution branches under proper label guidance to address this challenge. We provide in-depth analysis of our framework and introduce the cascade feature fusion unit to quickly achieve high-quality segmentation. Our system yields real-time inference on a single GPU card with decent quality results evaluated on challenging datasets like Cityscapes, CamVid and COCO-Stuff.

In this paper, we introduce a new model for leveraging unlabeled data to improve generalization performances of image classifiers: a two-branch encoder-decoder architecture called HybridNet. The first branch receives supervision signal and is dedicated to the extraction of invariant class-related representations. The second branch is fully unsupervised and dedicated to model information discarded by the first branch to reconstruct input data. To further support the expected behavior of our model, we propose an original training objective. It favors stability in the discriminative branch and complementarity between the learned representations in the two branches. HybridNet is able to outperform state-of-the-art results on CIFAR-10, SVHN and STL-10 in various semi-supervised settings. In addition, visualizations and ablation studies validate our contributions and the behavior of the model on both CIFAR-10 and STL-10 datasets.

We propose a light-weight and highly robust real-time human performance capture method based on a single depth camera and sparse inertial measurement units (IMUs). The proposed method combines non-rigid surface tracking and volumetric surface fusion to simultaneously reconstruct challenging motions, detailed geometries and the inner human body shapes of a clothed subject. The proposed hybrid motion tracking algorithm and efficient per-frame sensor calibration technique enable non-rigid surface reconstruction for fast motions and challenging poses with severe occlusions. Significant fusion artifacts are reduced using a new confidence measurement for our adaptive TSDF-based fusion. The above contributions benefit each other in our real-time reconstruction system, which enable practical human performance capture that is real-time, robust, low-cost and easy to deploy. Our experiments show how extremely challenging performances and loop closure problems can be handled successfully.