ECCV 2020 论文列表

Recent works have shown that exploiting unlabeled data through label propagation can substantially reduce the labeling cost, which has been a critical issue in developing visual recognition models. Yet, how to propagate labels reliably, especially on a dataset with unknown outliers, remains an open question. Conventional methods such as linear diffusion lack the capability of handling complex graph structures and may perform poorly when the seeds are sparse. Latest methods based on graph neural networks would face difficulties on performance drop as they scale out to noisy graphs. To overcome these difficulties, we propose a new framework that allows labels to be propagated reliably on large-scale real-world data. This framework incorporates (1) a local graph neural network to predict accurately on varying local structures while maintaining high scalability, and (2) a confidence-based path scheduler that identifies outliers and moves forward the propagation frontier in a prudent way. Experiments on both ImageNet and Ms-Celeb-1M show that our confidence guided framework can significantly improve the overall accuracies of the propagated labels, especially when the graph is very noisy."

In this work, we propose a method for training distributed GAN with sequential temporary discriminators. Our proposed method tackles the challenge of training GAN in the federated learning manner: How to update the generator with a flow of temporary discriminators? We apply our proposed method to learn a self-adaptive generator with a series of local discriminators from multiple data centers. We show our design of loss function indeed learns the correct distribution with provable guarantees. The empirical experiments show that our approach is capable of generating synthetic data which is practical for real-world applications such as training a segmentation model. Our TDGAN Code is available at: https://github.com/huiqu18/TDGAN-PyTorch."

Predicting the crowd behavior in complex environments is a key requirement for crowd and disaster management, architectural design, and urban planning. Given a crowd's immediate state, current approaches must be successively repeated over multiple time-steps for long-term predictions, leading to compute expensive and error-prone results. However, most applications require the ability to accurately predict hundreds of possible simulation outcomes (e.g., under different environment and crowd situations) at real-time rates, for which these approaches are prohibitively expensive. We propose the first deep framework to instantly predict the long-term flow of crowds in arbitrarily large, realistic environments. Central to our approach are a novel representation CAGE, which efficiently encodes crowd scenarios into compact, fixed-size representations that losslessly represent the environment, and a modified SegNet architecture for instant long-term crowd flow prediction. We conduct comprehensive experiments on novel synthetic and real datasets. Our results indicate that our approach is able to capture the essence of real crowd movement over very long time periods, while generalizing to never-before-seen environments and crowd contexts."

Graph matching has been a fundamental problem in computer vision and pattern recognition, for its practical flexibility as well as NP hardness challenge. Though the matching between two graphs and among multiple graphs have been intensively studied in literature, the online setting for incremental matching of a stream of graphs has been rarely considered. In this paper, we treat the graphs as graphs on a super-graph, and propose a novel breadth first search based method for expanding the neighborhood on the super-graph for a new coming graph, such that the matching with the new graph can be efficiently performed within the constructed neighborhood. Then depth first search is performed to update the overall pairwise matchings. Moreover, we show our approach can also be readily used in the batch mode setting, by adaptively determining the order of coming graph batch for matching, still under the neighborhood expansion based incremental matching framework. Extensive experiments on both online and offline matching of graph collections show our approach’s state-of-the-art accuracy and efficiency. The source code will be made public available."

Conditioning analysis uncovers the landscape of an optimization objective by exploring the spectrum of its curvature matrix. This has been well explored theoretically for linear models. We extend this analysis to deep neural networks (DNNs) in order to investigate their learning dynamics. To this end, we propose layer-wise conditioning analysis, which explores the optimization landscape with respect to each layer independently. Such an analysis is theoretically supported under mild assumptions that approximately hold in practice. Based on our analysis, we show that batch normalization (BN) can stabilize the training, but sometimes result in the false impression of a local minimum, which has detrimental effects on the learning. Besides, we experimentally observe that BN can improve the layer-wise conditioning of the optimization problem. Finally, we find that the last linear layer of a very deep residual network displays ill-conditioned behavior. We solve this problem by only adding one BN layer before the last linear layer, which achieves improved performance over the original and pre-activation residual networks."

Deep neural networks with a massive number of layers have made a remarkable breakthrough on single image super-resolution (SR), but sacrifice computation complexity and memory storage. To address this problem, we focus on the lightweight models for fast and accurate image SR. Due to the frequent use of residual block (RB) in SR models, we pursue an economical structure to adaptively combine RBs. Drawing lessons from lattice filter bank, we design the lattice block (LB) in which two butterfly structures are applied to combine two RBs. LB has the potential of various linear combinations of two RBs. Each case of LB depends on the combination coefficients which are determined by the attention mechanism. LB favors the lightweight SR model with the reduction of about half amount of the parameters while keeping the similar SR performance. Moreover, we propose a lightweight SR model, LatticeNet, which uses series connection of LBs and the backward feature fusion. Extensive experiments demonstrate that our proposal can achieve superior accuracy on four available benchmark datasets against other state-of-the-art methods, while maintaining relatively low computation and memory requirements."

In real-world applications, data are often associated with different labels. Although most extant multi-label learning algorithms consider the label correlations, they rarely consider the topic information hidden in the labels, where each topic is a group of related labels and different topics have different groups of labels. In our study, we assume that there exists a common feature representation for labels in each topic. Then, feature-label correlation can be exploited in the latent topic space. This paper shows that the sample and feature exaction, which are two important procedures for removing noisy and redundant information encoded in training samples in both sample and feature perspectives, can be effectively and efficiently performed in the latent topic space by considering topic-based feature-label correlation. Empirical studies on several benchmarks demonstrate the effectiveness and efficiency of the proposed topic-aware framework."

Zero-shot learning strives to classify unseen categories for which no data is available during training. In the generalized variant, the test samples can further belong to seen or unseen categories. The state-of-the-art relies on Generative Adversarial Networks that synthesize unseen class features by leveraging class-specific semantic embeddings. During training, they generate semantically consistent features, but discard this constraint during feature synthesis and classification. We propose to enforce semantic consistency at all stages of (generalized) zero-shot learning: training, feature synthesis and classification. We first introduce a feedback loop, from a semantic embedding decoder, that iteratively refines the generated features during both the training and feature synthesis stages. The synthesized features together with their corresponding latent embeddings from the decoder are then transformed into discriminative features and utilized during classification to reduce ambiguities among categories. Experiments on (generalized) zero-shot object and action classification reveal the benefit of semantic consistency and iterative feedback, outperforming existing methods on six zero-shot learning benchmarks. Source code at https://github.com/akshitac8/tfvaegan."

Prior work in visual dialog has focused on training deep neural models on VisDial in isolation. Instead, we present an approach to leverage pretraining on related vision-language datasets before transferring to visual dialog. We adapt the recently proposed ViLBERT model (Lu et al. 2019) for multi-turn visually-grounded conversations. Our model is pretrained on the Conceptual Captions and Visual Question Answering datasets, and finetuned on VisDial. Our best single model outperforms prior published work (including model ensembles) by more than 1% absolute on NDCG and MRR.

Large-scale few-shot learning aims at identifying hundreds of novel object categories where each category has only a few samples. It is a challenging problem since (1) the identifying process is susceptible to over-fitting with limited samples of an object, and (2) the sample imbalance between a base (known knowledge) category and a novel category is easy to bias the recognition results. To solve these problems, we propose a method based on multi-modal knowledge discovery. First, we use the visual knowledge to help the feature extractors focus on different visual parts. Second, we design a classifier to learn the distribution over all categories. In the second stage, we develop three schemes to minimize the prediction error and balance the training procedure: (1) Hard labels are used to provide precise supervision. (2) Semantic textual knowledge is utilized as weak supervision to find the potential relations between the novel and the base categories. (3) An imbalance control is presented from the data distribution to alleviate the recognition bias towards the base categories. We apply our method on three benchmark datasets, and it achieves state-of-the-art performances in all the experiments."

Video recognition has been advanced in recent years by benchmarks with rich annotations. However, research is still mainly limited to human action or sports recognition - focusing on a highly specific video understanding task and thus leaving a significant gap towards describing the overall content of a video. We fill this gap by presenting a large-scale ``Holistic Video Understanding Dataset""~(HVU). HVU is organized hierarchically in a semantic taxonomy that focuses on multi-label and multi-task video understanding as a comprehensive problem that encompasses the recognition of multiple semantic aspects in the dynamic scene. HVU contains approx.~572k videos in total with 9 million annotations for training, validation, and test set spanning over 3142 labels. HVU encompasses semantic aspects defined on categories of scenes, objects, actions, events, attributes, and concepts which naturally capture the real-world scenarios.

Most of existing object detectors usually adopt a small training batch size ( ~16), which severely hinders the whole community from exploring large-scale datasets due to the extremely long training procedure. In this paper, we propose a versatile large batch optimization framework for object detection, named LargeDet, which successfully scales the batch size to larger than 1K for the first time. Specifically, we present a novel Periodical Moments Decay LAMB (PMD-LAMB) algorithm to effectively reduce the negative effects of the lagging historical gradients. Additionally, the Synchronized Batch Normalization (SyncBN) is utilized to help fast convergence. With LargeDet, we can not only prominently shorten the training period, but also significantly improve the detection accuracy of sparsely annotated large-scale datasets. For instance, we can finish the training of ResNet50 FPN detector on COCO within 12 minutes. Moreover, we achieve 12.2% mAP@0.5 absolute improvement for ResNet50 FPN on Open Images by training with batch size 640. "

We introduce a solution to large scale Augmented Reality for outdoor scenes by registering camera images to textured Digital Elevation Models (DEMs). To accomodate the inherent differences in appearance between real images and DEMs, we train a cross-domain feature descriptor using Structure From Motion (SFM) reconstructions to acquire training data. Our method runs efficiently on a mobile device, and outperforms existing learned and hand designed feature descriptors for this task."

Deep implicit field regression methods are effective for 3D reconstruction from single-view images. However, the impact of different sampling patterns on the reconstruction quality is not well-understood. In this work, we first study the effect of point set discrepancy on the network training. Based on Farthest Point Sampling algorithm, we propose a sampling scheme that theoretically encourages better generalization performance, and results in fast convergence for SGD-based optimization algorithms. Secondly, based on the reflective symmetry of an object, we propose a feature fusion method that alleviates issues due to self-occlusions which makes it difficult to utilize local image features. Our proposed system Ladybird is able to create high quality 3D object reconstructions from a single input image. We evaluate Ladybird on a large scale 3D dataset (ShapeNet) demonstrating highly competitive results in terms of Chamfer distance, Earth Mover's distance and Intersection Over Union (IoU). "

In this paper we propose a new intermediate supervision method, named LabelEnc, to boost the training of object detection systems. The key idea is to introduce a novel label encoding function, mapping the ground-truth labels into latent embedding, acting as an auxiliary intermediate supervision to the detection backbone during training. Our approach mainly involves a two-step training procedure. First, we optimize the label encoding function via an AutoEncoder defined in the label space, approximating the ""desired"" intermediate representations for the target object detector. Second, taking advantage of the learned label encoding function, we introduce a new auxiliary loss attached to the detection backbones, thus benefiting the performance of the derived detector. Experiments show our method improves a variety of detection systems by around 2% on COCO dataset, no matter one-stage or two-stage frameworks. Moreover, the auxiliary structures only exist during training, i.e. it is completely cost-free in inference time."

Curriculum learning can improve neural network training by guiding the optimization to desirable optima. We propose a novel curriculum learning approach for image classification that adapts the loss function by changing the label representation. The idea is to use a probability distribution over classes as target label, where the class probabilities reflect the similarity to the true class. Gradually, this label representation is shifted towards the standard one-hot-encoding. That is, in the beginning minor mistakes are corrected less than large mistakes, resembling a teaching process in which broad concepts are explained first before subtle differences are taught. The class similarity can be based on prior knowledge. For the special case of the labels being natural words, we propose a generic way to automatically compute the similarities. The natural words are embedded into Euclidean space using a standard word embedding. The probability of each class is then a function of the cosine similarity between the vector representations of the class and the true label. The proposed label-similarity curriculum learning (LCL) approach was empirically evaluated using several popular deep learning architectures for image classification tasks applied to five datasets including ImageNet, CIFAR100, and AWA2. In all scenarios, LCL was able to improve the classification accuracy on the test data compared to standard training. Code to reproduce results is available at https://github.com/speedystream/LCL."

The problems of shape classification and part segmentation from 3D point clouds have garnered increasing attention in the last few years. Both of these problems, however, suffer from relatively small training sets, creating the need for statistically efficient methods to learn 3D shape representations. In this paper, we investigate the use of Approximate Convex Decompositions (ACD) as a self-supervisory signal for label-efficient learning of point cloud representations. We show that using ACD to approximate ground truth segmentation provides excellent self-supervision for learning 3D point cloud representations that are highly effective on downstream tasks. We report improvements over the state-of-the-art for unsupervised representation learning on the ModelNet40 shape classification dataset and significant gains in few-shot part segmentation on the ShapeNetPart dataset. Our source code is publicly available."

Unsupervised domain adaptation enables to alleviate the need for pixel-wise annotation in the semantic segmentation. One of the most common strategies is to translate images from the source domain to the target domain and then align their marginal distributions in the feature space using adversarial learning. However, source-to-target translation enlarges the bias in translated images and introduces extra computations, owing to the dominant data size of the source domain. Furthermore, consistency of the joint distribution in source and target domains cannot be guaranteed through global feature alignment. Here, we present an innovative framework, designed to mitigate the image translation bias and align cross-domain features with the same category. This is achieved by 1) performing the target-to-source translation and 2) reconstructing both source and target images from their predicted labels. Extensive experiments on adapting from synthetic to real urban scene understanding demonstrate that our framework competes favorably against existing state-of-the-art methods."

Motivated by the problem relatedness between unsupervised domain adaptation (UDA) and semi-supervised learning (SSL), many state-of-the-art UDA methods adopt SSL principles (e.g., the cluster assumption) as their learning ingredients. However, they tend to overlook the very domain-shift nature of UDA. In this work, we take a step further to study the proper extensions of SSL techniques for UDA. Taking the algorithm of label propagation (LP) as an example, we analyze the challenges of adopting LP to UDA and theoretically analyze the conditions of affinity graph/matrix construction in order to achieve better propagation of true labels to unlabeled instances. Our analysis suggests a new algorithm of Label Propagation with Augmented Anchors (A$^2$LP), which could potentially improve LP via generation of unlabeled virtual instances (i.e., the augmented anchors) with high-confidence label predictions. To make the proposed A$^2$LP useful for UDA, we propose empirical schemes to generate such virtual instances. The proposed schemes also tackle the domain-shift challenge of UDA by alternating between pseudo labeling via A$^2$LP and domain-invariant feature learning. Experiments show that such a simple SSL extension improves over representative UDA methods of domain-invariant feature learning, and could empower two state-of-the-art methods on benchmark UDA datasets. Our results show the value of further investigation on SSL techniques for UDA problems."