ECCV 2020 论文列表

In this paper, we address the problem of nighttime defogging from a single image. We propose a framework consisting of two main modules: grayscale and color modules. Given an RGB foggy nighttime image, our grayscale module takes the grayscale version of the image as input, and decomposes it into high and low frequency layers. The high frequency layer contains the scene texture information, which is less affected by fog; and, the low frequency layer contains the scene layout/structure information including fog and glow. Our grayscale module then enhances the visibility of the textures in the high frequency layers, and removes the presence of glow and fog in the low frequency layers. Having processed the high/low frequency information, it fuses the layers to obtain a grayscale defogged image. Our second module, the color module, takes the original RGB image as input. The module has similar operations to those of the grayscale module. However, to obtain fog-free high and low frequency information, the module is guided by the grayscale module. The reason of doing this is because grayscale images are less affected by multiple colors of atmospheric light, which are commonly present in nighttime scenes. Moreover, having the grayscale module allows us to have consistency losses between the outputs of the two modules, which is critical to our framework, since we do not have paired ground-truths for our real data. Our extensive experiments on real foggy nighttime images show the effectiveness of our method."

The introduction of non-local blocks to the traditional CNN architecture enhances its performance for various computer vision tasks by improving its capabilities of capturing long range dependencies. However, the usage of non-local blocks may also introduce new threats to computer vision systems. Therefore, it is important to study the threats caused by non-local blocks before directly applying them to commercial systems. In this paper, two new threats named disappearing attack and appearing attack against object detectors with a non-local block are investigated. The former aims at misleading an object detector with a non-local block such that it is unable to detect a target object category while the latter aims at misleading the object detector such that it detects a predefined object category, which is not present in images. Different from the existing attacks against object detectors, these threats are able to be performed in long range cases. This means that the target object and the universal adversarial patches learned from the proposed algorithms can have long distance between them. To examine the threats, digital and physical experiments are conducted on Faster R-CNN with a non-local block and 6331 images from 56 videos. The experiments show that the universal patches are able to mislead the detector with greater probabilities. To explain the threats from non-local blocks, the reception fields of CNN models with and without non-local blocks are studied empirically and theoretically."

In architecture and computer-aided design, wireframes (i.e., line-based models) are widely used as basic 3D models for design evaluation and fast design iterations. However, unlike a full design file, a wireframe model lacks critical information, such as detailed shape, texture, and materials, needed by a conventional renderer to produce 2D renderings of the objects or scenes. In this paper, we bridge the information gap by generating photo-realistic rendering of indoor scenes from wireframe models in an image translation framework. While existing image synthesis methods can generate visually pleasing images for common objects such as faces and birds, these methods do not explicitly model and preserve essential structural constraints in a wireframe model, such as junctions, parallel lines, and planar surfaces. To this end, we propose a novel model based on a structure-appearance joint representation learned from both images and wireframes. In our model, structural constraints are explicitly enforced by learning a joint representation in a shared encoder network that must support the generation of both images and wireframes. Experiments on a wireframe-scene dataset show that our wireframe-to-image translation model significantly outperforms the state-of-the-art methods in both visual quality and structural integrity of generated images."

We present Neural Voice Puppetry, a novel approach for audio-driven facial video synthesis. Given an audio sequence of a source person or digital assistant, we generate a photo-realistic output video of a target person that is in sync with the audio of the source input. This audio-driven facial reenactment is driven by a deep neural network that employs a latent 3D face model space. Through the underlying 3D representation, the model inherently learns temporal stability while we leverage neural rendering to generate photo-realistic output frames. Our approach generalizes across different people, allowing us to synthesize videos of a target actor with the voice of any unknown source actor or even synthetic voices that can be generated utilizing standard text-to-speech approaches. Neural Voice Puppetry has a variety of use-cases, including audio-driven video avatars, video dubbing, and text-driven video synthesis of a talking head. We demonstrate the capabilities of our method in a series of audio- and text-based puppetry examples, including comparisons to state-of-the-art techniques and a user study."

Human re-rendering from a single image is a starkly under-constrained problem and state-of-the-art algorithms often exhibit un-desired artefacts, such as oversmoothing, unrealistic distortions of thebody parts and garments, or implausible changes of the texture. To ad-dress these challenges, we propose a new method for neural re-renderingof a human under a novel user-defined pose and viewpoint given oneinput image. Our algorithm represents body pose and shape as a para-metric mesh which can be reconstructed from a single image and easilyreposed. Instead of a color-based UV texture-map, our approach furtheremploys a learned high-dimensional UV feature-map to encode appear-ance. This rich implicit representation captures detailed appearance vari-ation across poses, viewpoints, person identities and clothing styles bet-ter than learned color texture maps. The body model with the renderedfeature-maps is fed through a neural image-translation network that cre-ates the final rendered color image. The above components are combinedin an end-to-end-trained neural network architecture that takes as in-put a source person image, and images of the parametric body modelin the source pose and desired target pose. Experimental evaluationdemonstrates that our approach produces higher quality single-imagere-rendering results than existing methods."

Neural Architecture Search methods are effective but often use complex algorithms to come up with the best architecture. We propose an approach with three basic steps that is conceptually much simpler. First we train $N$ random architectures to generate $N$ (architecture, validation accuracy) pairs and use them to train a regression model that predicts accuracies for architectures. Next, we use this regression model to predict the validation accuracies of a large number of random architectures. Finally, we train the top-$K$ predicted architectures and deploy the model with the best validation result. While this approach seems simple, it is more than $20 imes$ as sample efficient as Regularized Evolution on the NASBench-101 benchmark. On ImageNet, it approaches the efficiency of more complex and restrictive approaches based on weight sharing such as ProxylessNAS while being fully (embarrassingly) parallelizable and friendly to hyper-parameter tuning."

We present a new point-based approach for modeling the appearance of real scenes. The approach uses a raw point cloud as the geometric representation of a scene, and augments each point with a learnable neural descriptor that encodes local geometry and appearance. A deep rendering network is learned in parallel with the descriptors, so that new views of the scene can be obtained by passing the rasterizations of a point cloud from new viewpoints through this network. The input rasterizations use the learned descriptors as point pseudo-colors. We show that the proposed approach can be used for modeling complex scenes and obtaining their photorealistic views, while avoiding explicit surface estimation and meshing. In particular, compelling results are obtained for scenes scanned using hand-held commodity RGB-D sensors as well as standard RGB cameras even in the presence of objects that are challenging for standard mesh-based modeling."

Recent methods for 6D pose estimation of objects assume either textured 3D models or real images that cover the entire range of target poses. However, it is difficult to obtain textured 3D models and annotate the poses of objects in real scenarios. This paper proposes a method, Neural Object Learning (NOL), that creates synthetic images of objects in arbitrary poses by combining only a few observations from cluttered images. A novel refinement step is proposed to align inaccurate poses of objects in source images, which results in better quality images. Evaluations performed on two public datasets show that the rendered images created by NOL lead to state-of-the-art performance in comparison to methods that use 13 times the number of real images. Evaluations on our new dataset show multiple objects can be trained and recognized simultaneously using a sequence of a fixed scene. "

In this paper, we propose a generic neural-based hair rendering pipeline that can synthesize photo-realistic images from virtual 3D hair models. Unlike existing supervised translation methods that require model-level similarity to preserve consistent structure representation for both real images and fake renderings, our method adopts an unsupervised solution to work on arbitrary hair models. The key component of our method is a shared latent space to encode appearance-invariant structure information of both domains, which generates realistic renderings conditioned by extra appearance inputs. This is achieved by domain-specific pre-disentangled structure representation, partially shared domain encoder layers and a structure discriminator. We also propose a simple yet effective temporal conditioning method to enforce consistency for video sequence generation. We demonstrate the superiority of our method by testing it on a large number of portraits and comparing it with alternative baselines and state-of-the-art unsupervised image translation methods."

We propose a neural geometric parser learning single image camera calibration for man-made scenes. Unlike previous neural approaches that rely only on semantic cues obtained from neural networks, our approach considers both semantic and geometric cues, resulting in significant accuracy improvement. The proposed framework consists of two networks. Using line segments of an image as geometric cues, the first network estimates the zenith vanishing point and generates several candidates consisting of the camera rotation and focal length. The second network evaluates each candidate based on the given image and the geometric cues, where prior knowledge of man-made scenes is used for the evaluation. With the supervision of datasets consisting of the horizontal line and focal length of the images, our networks can be trained to estimate the same camera parameters. Based on the Manhattan world assumption, we can further estimate the camera rotation and focal length in a weakly supervised manner. The experimental results reveal that the performance of our neural approach is significantly higher than that of existing state-of-the-art camera calibration techniques for single images of indoor and outdoor scenes."

Graphic design is essential for visual communication with layouts being fundamental to composing attractive designs. Layout generation differs from pixel-level image synthesis and is unique in terms of the requirement of mutual relations among the desired components. We propose a method for design layout generation that can satisfy user-specified constraints. The proposed neural design network (NDN) consists of three modules. The first module predicts a graph with complete relations from a graph with user-specified relations. The second module generates a layout from the predicted graph. Finally, the third module fine-tunes the predicted layout. Quantitative and qualitative experiments demonstrate that the generated layouts are visually similar to real design layouts. We also construct real designs based on predicted layouts for a better understanding of the visual quality. Finally, we demonstrate a practical application on layout recommendation."

We introduce the first dense neural non-rigid structure from motion (N-NRSfM) approach, which can be trained end-to-end in an unsupervised manner from 2D point tracks. Compared to the competing methods, our combination of loss functions is fully-differentiable and can be readily integrated into deep-learning systems. We formulate the deformation model by an auto-decoder and impose subspace constraints on the recovered latent space function in a frequency domain. Thanks to the state recurrence cue, we classify the reconstructed non-rigid surfaces based on their similarity and recover the period of the input sequence. Our N-NRSfM approach achieves competitive accuracy on widely-used benchmark sequences and high visual quality on various real videos. Apart from being a standalone technique, our method enables multiple applications including shape compression, completion and interpolation, among others. Combined with an encoder trained directly on 2D images, we perform scenario-specific monocular 3D shape reconstruction at interactive frame rates. To facilitate the reproducibility of the results and boost the new research direction, we open-source our code and provide trained models for research purposes. "

We are interested in the detection and segmentation of anomalies in images where the anomalies are typically small (i.e., a small tear in woven fabric, broken pin of an IC chip). From a statistical learning point of view, anomalies have low occurrence probability and are not from the main modes of a data distribution. Learning a generative model of anomalous data from a natural distribution of data can be difficult because the data distribution is heavily skewed towards a large amount of non-anomalous data. When training a generative model on such imbalanced data using an iterative learning algorithm like stochastic gradient descent (SGD), we observe an expected yet interesting trend in the loss values (a measure of the learned models performance) after each gradient update across data samples. Naturally, as the model sees more non-anomalous data during training, the loss values over a non-anomalous data sample decreases, while the loss values on an anomalous data sample fluctuates. In this work, our key hypothesis is that this change in loss values during training can be used as a feature to identify anomalous data. In particular, we propose a novel semi-supervised learning algorithm for anomaly detection and segmentation using an anomaly classifier that uses as input the extit{loss profile} of a data sample processed through an autoencoder. The loss profile is defined as a sequence of reconstruction loss values produced during iterative training. To amplify the difference in loss profiles between anomalous and non-anomalous data, we also introduce a Reinforcement Learning based meta-algorithm, which we call the neural batch sampler, to strategically sample training batches during autoencoder training. Experimental results on multiple datasets with a high diversity of textures and objects, often with multiple modes of defects within them, demonstrate the capabilities and effectiveness of our method when compared with existing state-of-the-art baselines."

Multiple rotation averaging is an essential task for structure from motion, mapping, and robot navigation. The task is to estimate the absolute orientations of several cameras given some of their noisy relative orientation measurements. The conventional methods for this task seek parameters of the absolute orientations that agree best with the observed noisy measurements according to a robust cost function. These robust cost functions are highly nonlinear and are designed based on certain assumptions about the noise and outlier distributions. In this work, we aim to build a neural network that learns the noise patterns from the data and predict/regress the model parameters from the noisy relative orientations. The proposed network is a combination of two networks: (1) a view-graph cleaning network, which detects outlier edges in the view-graph and rectifies noisy measurements; and (2) a fine-tuning network, which fine-tunes an initialization of absolute orientations bootstrapped from the cleaned graph, in a single step. The proposed combined network is very fast, moreover, being trained on a large number of synthetic graphs, it is more accurate than the conventional iterative optimization methods. Although the idea of replacing robust optimization methods by a graph-based network is demonstrated only for multiple rotation averaging, it could easily be extended to other graph-based geometric problems, for example, pose-graph optimization. "

In pathological diagnosis, since the proportion of the adenocarcinoma subtypes is related to the recurrence rate and the survival time after surgery, the proportion of cancer subtypes for pathological images has been recorded as diagnostic information in some hospitals. In this paper, we propose a subtype segmentation method that uses such proportional labels as weakly supervised labels. If the estimated class rate is higher than that of the annotated class rate, we generate negative pseudo labels, which indicate, ``input image does not belong to this negative label,'' in addition to standard pseudo labels. It can force out the low confidence samples and mitigate the problem of positive pseudo label learning which cannot label low confident unlabeled samples. Our method outperformed the state-of-the-art semi-supervised learning (SSL) methods."

In this paper, we unconventionally propose to adopt appropriate negative-margin to softmax loss for few-shot classification, which surprisingly works well for the open-set scenarios of few-shot classification. We then provide the intuitive explanation and the theoretical proof to understand why negative margin works well for few-shot classification. This claim is also demonstrated via sufficient experiments. With the negative-margin softmax loss, our approach achieves the state-of-the-art performance on all three standard benchmarks of few-shot classification. In the future, the negative margin may be applied in more general open-set scenarios that do not restrict the number of samples in novel classes."

We present a method that achieves state-of-the-art results for synthesizing novel views of complex scenes by optimizing an underlying continuous volumetric scene function using a sparse set of input views. Our algorithm represents a scene using a fully-connected (non-convolutional) deep network, whose input is a single continuous 5D coordinate (spatial location $(x,y,z)$ and viewing direction $( heta,\phi)$) and whose output is the volume density and view-dependent emitted radiance at that spatial location. We synthesize views by querying 5D coordinates along camera rays and use classic volume rendering techniques to project the output colors and densities into an image. Because volume rendering is naturally differentiable, the only input required to optimize our representation is a set of images with known camera poses. We describe how to effectively optimize neural radiance fields to render photorealistic novel views of scenes with complicated geometry and appearance, and demonstrate results that outperform prior work on neural rendering and view synthesis. View synthesis results are best viewed as videos, so we urge readers to view our supplementary video for convincing comparisons. We will make our code and data available upon publication."

Supervised learning in large discriminative models is a mainstay for modern computer vision. Such an approach necessitates investing in large-scale human-annotated datasets for achieving state-of-the-art results. In turn, the efficacy of supervised learning may be limited by the size of the human annotated dataset. This limitation is particularly notable for image segmentation tasks, where the expense of human annotation is especially large, yet large amounts of unlabeled data may exist. In this work, we ask if we may leverage semi-supervised learning in unlabeled video sequences and extra images to improve the performance on urban scene segmentation, simultaneously tackling semantic, instance, and panoptic segmentation. The goal of this work is to avoid the construction of sophisticated, learned architectures specific to label propagation (e.g., patch matching and optical flow). Instead, we simply predict pseudo-labels for the unlabeled data and train subsequent models with both human-annotated and pseudo-labeled data. The procedure is iterated for several times. As a result, our Naive-Student model, trained with such simple yet effective iterative semi-supervised learning, attains state-of-the-art results at all three Cityscapes benchmarks, reaching the performance of 67.8% PQ, 42.6% AP, and 85.2% mIOU on the test set. We view this work as a notable step towards building a simple procedure to harness unlabeled video sequences and extra images to surpass state-of-the-art performance on core computer vision tasks."

In this paper, we propose an efficient NAS algorithm for generating task-specific models that are competitive under multiple competing objectives. It comprises of two surrogates, one at the architecture level to improve sample efficiency and one at the weights level, through a supernet, to improve gradient descent training efficiency. On standard benchmark datasets (C10, C100, ImageNet), the resulting models, dubbed NSGANetV2, either match or outperform models from existing approaches with the search being orders of magnitude more sample efficient. Furthermore, we demonstrate the effectiveness and versatility of the proposed method on six diverse non-standard datasets, e.g. STL-10, Flowers102, Oxford Pets, FGVC Aircrafts etc. In all cases, NSGANetV2s improve the state-of-the-art (under mobile setting), suggesting that NAS can be a viable alternative to conventional transfer learning approaches in handling diverse scenarios such as small-scale or fine-grained datasets. Code is available at https://github.com/mikelzc1990/nsganetv2"