ECCV 2018 Papers

Despite the large number of both commercial and academic methods for Automatic License Plate Recognition (ALPR), most existing approaches are focused on a specific license plate (LP) region (e.g. European, US, Brazilian, Taiwanese, etc.), and frequently explore datasets containing approximately frontal images. This work proposes a complete ALPR system focusing on unconstrained capture scenarios, where the LP might be considerably distorted due to oblique views. Our main contribution is the introduction of a novel Convolutional Neural Network (CNN) capable of detecting and rectifying multiple distorted license plates in a single image, which are fed to an Optical Character Recognition (OCR) method to obtain the final result. As an additional contribution, we also present manual annotations for a challenging set of LP images from different regions and acquisition conditions. Our experimental results indicate that the proposed method, without any parameter adaptation or fine tuning for a specific scenario, performs similarly to state-of-the-art commercial systems in traditional datasets, and outperforms both academic and commercial approaches in challenging datasets.

Despite many advances in deep-learning based semantic segmentation, performance drop due to distribution mismatch is often encountered in the real world. Recently, a few domain adaptation and active learning approaches have been proposed to mitigate the performance drop. However, very little attention has been made toward leveraging information in videos which are naturally captured in most camera systems. In this work, we propose to leverage "motion prior" in videos for improving human segmentation in a weakly-supervised active learning setting. By extracting motion information using optical flow in videos, we can extract candidate foreground motion segments (referred to as motion prior) potentially corresponding to human segments. We propose to learn a memory-network-based policy model to select strong candidate segments (referred to as strong motion prior) through reinforcement learning. The selected segments have high precision and are directly used to finetune the model. In a newly collected surveillance camera dataset and a publicly available UrbanStreet dataset, our proposed method improves the performance of human segmentation across multiple scenes and modalities (i.e., RGB to Infrared (IR)). Last but not least, our method is empirically complementary to existing domain adaptation approaches such that additional performance gain is achieved by combining our weakly-supervised active learning approach with domain adaptation approaches.

In video captioning task, the best practice has been achieved by attention-based models which associate salient visual components with sentences in the video. However, existing study follows a common procedure which includes a frame-level appearance modeling and motion modeling on equal interval frame sampling, which may bring about redundant visual information, sensitivity to content noise and unnecessary computation cost. We propose a plug-and-play PickNet to perform informative frame picking in video captioning. Based on a standard encoder-decoder framework, we develop a reinforcement-learning-based procedure to train the network sequentially, where the reward of each frame picking action is designed by maximizing visual diversity and minimizing discrepancy between generated caption and the ground-truth. The rewarded candidate will be selected and the corresponding latent representation of encoder-decoder will be updated for future trials. This procedure goes on until the end of the video sequence. Consequently, a compact frame subset can be selected to represent the visual information and perform video captioning without performance degradation. Experiment results show that our model can achieve competitive performance across popular benchmarks while only 6~8 frames are used.

Video motion magnification techniques allow us to see small motions previously invisible to the naked eyes, such as those of vibrating airplane wings, or swaying buildings under the influence of the wind. Because the motion is small, the magnification results are prone to noise or excessive blurring. The state of the art relies on hand-designed filters to extract representations that may not be optimal. In this paper, we seek to learn the filters directly from examples using deep convolutional neural networks. To make training tractable, we carefully design a synthetic dataset that captures small motion well, and use two-frame input for training. We show that the learned filters achieve high-quality results on real videos, with less ringing artifacts and better noise characteristics than previous methods. While our model is not trained with temporal filters, we found that the temporal filters can be used with our extracted representations up to a moderate magnification, enabling a frequency-based motion selection. Finally, we analyze the learned filters and show that they behave similarly to the derivative filters used in previous works. Our code, trained model, and datasets will be available online.

In this paper, we study the classification problem in which we have access to easily obtainable surrogate for true labels, namely complementary labels, which specify classes that observations do extbf{not} belong to. Let $Y$ and $ar{Y}$ be the true and complementary labels, respectively. We first model the annotation of complementary labels via transition probabilities $P(ar{Y}=i|Y=j), i eq jin{1,cdots,c}$, where $c$ is the number of classes. Previous methods implicitly assume that $P(ar{Y}=i|Y=j), forall i eq j$, are identical, which is not true in practice because humans are biased toward their own experience. For example, as shown in Figure ef{complementary_label_cases}, if an annotator is more familiar with monkeys than prairie dogs when providing complementary labels for meerkats, she is more likely to employ ``monkey'' as a complementary label. We therefore reason that the transition probabilities will be different. In this paper, we propose a framework that contributes three main innovations to learning with extbf{biased} complementary labels: (1) It estimates transition probabilities with no bias. (2) It provides a general method to modify traditional loss functions and extends standard deep neural network classifiers to learn with biased complementary labels. (3) It theoretically ensures that the classifier learned with complementary labels converges to the optimal one learned with true labels. Comprehensive experiments on several benchmark datasets validate the superiority of our method to current state-of-the-art methods.

We introduce a saliency-based distortion layer for convolutional neural networks that helps to improve the spatial sampling of input data for a given task. Our differentiable layer can be added as a preprocessing block to existing task networks and trained altogether in an end-to-end fashion. The effect of the layer is to efficiently estimate how to sample from the original data in order to boost task performance. For example, for an image classification task in which the original data might range in size up to several megapixels, but where the desired input images to the task network are much smaller, our layer learns how best to sample from the underlying high resolution data in a manner which preserves task-relevant information better than uniform downsampling. This has the effect of creating distorted, caricature-like intermediate images, in which idiosyncratic elements of the image that improve task performance are zoomed and exaggerated. Unlike alternative approaches such as spatial transformer networks, our proposed layer is inspired by image saliency, computed efficiently from uniformly downsampled data, and degrades gracefully to a uniform sampling strategy under uncertainty. We apply our layer to improve existing networks for the tasks of human gaze estimation and fine-grained object classification. Code for our method is available in: http://github.com/recasens/Saliency-Sampler

Sum-of-squares objective functions are very popular in computer vision algorithms. However, these objective functions are not always easy to optimize. The underlying assumptions made by solvers are often not satisfied and many problems are inherently ill-posed. In this paper, we propose a neural nonlinear least squares optimization algorithm which learns to effectively optimize these cost functions even in the presence of adversities. Unlike traditional approaches, the proposed solver requires no hand-crafted regularizers or priors as these are implicitly learned from the data. We apply our method to the problem of motion stereo ie. jointly estimating the motion and scene geometry from pairs of images of a monocular sequence. We show that our learned optimizer is able to efficiently and effectively solve this challenging optimization problem.

Perceiving a scene most fully requires all the senses. Yet modeling how objects look and sound is challenging: most natural scenes and events contain multiple objects, and the audio track mixes all the sound sources together. We propose to learn audio-visual object models from unlabeled video, then exploit the visual context to perform audio source separation in novel videos. Our approach relies on a deep multi-instance multi-label learning framework to disentangle the audio frequency bases that map to individual visual objects, even without observing/hearing those objects in isolation. We show how the recovered disentangled bases can be used to guide audio source separation to obtain better-separated, object-level sounds. Our work is the first to learn audio source separation from large-scale "in the wild" videos containing multiple audio sources per video. We obtain state-of-the-art results on visually-aided audio source separation and audio denoising. Our video results: http://vision.cs.utexas.edu/projects/separating_object_sounds/

This paper presents a weakly-supervised approach to object instance segmentation. Starting with known or predicted object bounding boxes, we learn object masks by playing a game of cut-and-paste in an adversarial learning setup. A mask generator takes a detection box and Faster R-CNN features, and constructs a segmentation mask that is used to cut-and-paste the object into a new image location. The discriminator tries to distinguish between real objects, and those cut and pasted via the generator, giving a learning signal that leads to improved object masks. We verify our method experimentally using Cityscapes, COCO, and aerial image datasets, learning to segment objects without ever having seen a mask in training. Our method exceeds the performance of existing weakly supervised methods, without requiring hand-tuned segment proposals, and reaches 90% of supervised performance.

We present a data-driven approach to reconstructing high-resolution and detailed volumetric representations of 3D shapes. Although well studied, algorithms for volumetric fusion from multi-view depth scans are still prone to scanning noise and occlusions, making it hard to obtain high-fidelity 3D reconstructions. In this paper, inspired by recent advances in efficient 3D deep learning techniques, we introduce a novel cascaded 3D convolutional network architecture, which learns to reconstruct implicit surface representations from noisy and incomplete depth maps in a progressive, coarse-to-fine manner. To this end, we also develop an algorithm for end-to-end training of the proposed cascaded structure. Qualitative and quantitative experimental results on both simulated and real-world datasets demonstrate that the presented approach outperforms existing state-of-the-art work in terms of quality and fidelity of reconstructed models.

Recent methods for boundary or edge detection built on Deep Convolutional Neural Networks (CNNs) typically suffer from the issue of predicted edges being thick and need post-processing to obtain crisp boundaries. Highly imbalanced categories of boundary versus background in training data is one of main reasons for the above problem. In this work, the aim is to make CNNs produce sharp boundaries without post-processing. We introduce a novel loss for boundary detection, which is very effective for classifying imbalanced data and allows CNNs to produce crisp boundaries. Moreover, we propose an end-to-end network which adopts the bottom-up/top-down architecture to tackle the task. The proposed network effectively leverages hierarchical features and produces pixel-accurate boundary mask, which is critical to reconstruct the edge map. Our experiments illustrate that directly making crisp prediction not only promotes the visual results of CNNs, but also achieves better results against the state-of-the-art on the BSDS500 dataset (ODS F-score of .815) and the NYU Depth dataset (ODS F-score of .762).

Fine-grained classification is challenging due to the difficulty of finding discriminative features. Finding those subtle traits that fully characterize the object is not straightforward. To handle this circumstance, we propose a novel self-supervision mechanism to effectively localize informative regions without the need of fine-grained bounding-box/part annotations. Our model, termed NTS-Net for Navigator-Teacher-Scrutinizer Network, consists of a Navigator agent, a Teacher agent and a Scrutinizer agent. In consideration of intrinsic consistency between informativeness of the regions and their probability being ground-truth class, we design a novel training paradigm, which enables Navigator to detect most informative regions under the guidance from Teacher. After that, the Scrutinizer scrutinizes the proposed regions from Navigator and makes predictions. Our model can be viewed as a multi-agent cooperation, wherein agents benefit from each other, and make progress together. NTS-Net can be trained end-to-end, while provides accurate fine-grained classification predictions as well as highly informative regions during inference. We achieve state-of-the-art performance in extensive benchmark datasets.

Given a single RGB image of a complex outdoor road scene in the perspective view, we address the novel problem of estimating an occlusion-reasoned semantic scene layout in the top-view. This challenging problem not only requires an accurate understanding of both the 3D geometry and the semantics of the visible scene, but also of occluded areas. We propose a convolutional neural network that learns to predict occluded portions of the scene layout by looking around foreground objects like cars or pedestrians. But instead of hallucinating RGB values, we show that directly predicting the semantics and depths in the occluded areas enables a better transformation into the top-view. We further show that this initial top-view representation can be significantly enhanced by learning priors and rules about typical road layouts from simulated or, if available, map data. Crucially, training our model does not require costly or subjective human annotations for occluded areas or the top-view, but rather uses readily available annotations for standard semantic segmentation in the perspective view. We extensively evaluate and analyze our approach on the KITTI and Cityscapes data sets.

Semi-Global Matching (SGM) uses an aggregation scheme to combine costs from multiple 1D scanline optimizations that tends to hurt its accuracy in difficult scenarios. We propose replacing this aggregation scheme with a new learning-based method that fuses disparity proposals estimated using scanline optimization. Our proposed SGM-Forest algorithm solves this problem using per-pixel classification. SGM-Forest currently ranks 1st on the ETH3D stereo benchmark and is ranked competitively on the Middlebury 2014 and KITTI 2015 benchmarks. It consistently outperforms SGM in challenging settings and under difficult training protocols that demonstrate robust generalization, while adding only a small computational overhead to SGM.

We consider the problem of image-to-video translation, where an input image is translated into an output video containing motions of a single object. Recent methods for such problems typically train transformation networks to generate future frames conditioned on the structure sequence. Parallel work has shown that short high-quality motions can be generated by spatiotemporal generative networks that leverage temporal knowledge from the training data. We combine the benefits of both approaches and propose a two-stage generation framework where videos are generated from structures and then refined by temporal signals. To model motions more efficiently, we train networks to learn residual motion between the current and future frames, which avoids learning motion-irrelevant details. We conduct extensive experiments on two image-to-video translation tasks: facial expression retargeting and human pose forecasting. Superior results over the state-of-the-art methods on both tasks demonstrate the effectiveness of our approach.

The bullet-time effect, presented in feature film ``The Matrix", has been widely adopted in feature films and TV commercials to create an amazing stopping-time illusion. Producing such visual effects, however, typically requires using a large number of cameras/images surrounding the subject. In this paper, we present a learning-based solution that is capable of producing the bullet-time effect from only a small set of images. Specifically, we present a view morphing framework that can synthesize smooth and realistic transitions along extit{a circular view path} using as few as three reference images. We apply a novel cyclic rectification technique to align the reference images onto a common circle and then feed the rectified results into a deep network to predict its motion field and per-pixel visibility for new view interpolation. Comprehensive experiments on synthetic and real data show that our new framework outperforms the state-of-the-art and provides an inexpensive and practical solution for producing the bullet-time effects.

Multi-People Tracking in an open-world setting requires a special effort in precise detection. Moreover, temporal continuity in the detection phase gains more importance when scene cluttering introduces the challenging problems of occluded targets. For the purpose, we propose a deep network architecture that jointly extracts people body parts and associates them across short temporal spans. Our model explicitly deals with occluded body parts, by hallucinating plausible solutions of not visible joints. We propose a new end-to-end architecture composed by four branches (visible heatmaps, occluded heatmaps, part affinity fields and temporal affinity fields) fed by a time linker feature extractor. To overcome the lack of surveillance data with tracking, body part and occlusion annotations we created the vastest Computer Graphics dataset for people tracking in urban scenarios by exploiting a photorealistic videogame. It is up to now the vastest dataset (about 500.000 frames, almost 10 million body poses) of human body parts for people tracking in urban scenarios. Our architecture trained on virtual data exhibits good generalization capabilities also on public real tracking benchmarks, when image resolution and sharpness are high enough, producing reliable tracklets useful for further batch data association or re-id modules.

We propose a learning-based framework for acquiring a light field through a coded aperture camera. Acquiring a light field is a challenging task due to the amount of data. To make the acquisition process efficient, coded aperture cameras were successfully adopted; using these cameras, a light field is computationally reconstructed from several images that are acquired with different aperture patterns. However, it is still difficult to reconstruct a high-quality light field from only a few acquired images. To tackle this limitation, we formulated the entire pipeline of light field acquisition from the perspective of an auto-encoder. This auto-encoder was implemented as a stack of fully convolutional layers and was trained end-to-end by using a collection of training samples. We experimentally show that our method can successfully learn good image-acquisition and reconstruction strategies. With our method, light fields consisting of 5 x 5 or 8 x 8 images can be successfully reconstructed only from a few acquired images. Moreover, our method achieved superior performance over several state-of-the-art methods. We also applied our method to a real prototype camera to show that it is capable of capturing a real 3-D scene.

Photo blending is a common technique to create aesthetically pleasing artworks by combining multiple photos. However, the process of photo blending is usually time-consuming, and care must be taken in the process of blending, filtering, positioning, and masking each of the source photos. To make photo blending accessible to general public, we propose an efficient approach for automatic photo blending via deep learning. Specifically, given a foreground image and a background image, our proposed method automatically generates a set of blending photos with scores that indicate the aesthetics quality with the proposed quality network and policy network. Experimental results show that the proposed approach can effectively generate high quality blending photos with efficiency.