CVPR 2019 论文列表

State-of-the-art object detectors are usually trained on public datasets. They often face substantial difficulties when applied to a different domain, where the imaging condition differs significantly and the corresponding annotated data are unavailable (or expensive to acquire). A natural remedy is to adapt the model by aligning the image representations on both domains. This can be achieved, for example, by adversarial learning, and has been shown to be effective in tasks like image classification. However, we found that in object detection, the improvement obtained in this way is quite limited. An important reason is that conventional domain adaptation methods strive to align images as a whole, while object detection, by nature, focuses on local regions that may contain objects of interest. Motivated by this, we propose a novel approach to domain adaption for object detection to handle the issues in "where to look" and "how to align". Our key idea is to mine the discriminative regions, namely those that are directly pertinent to object detection, and focus on aligning them across both domains. Experiments show that the proposed method performs remarkably better than existing methods with about 4% 6% improvement under various domain-shift scenarios while keeping good scalability.

The ability to categorize is a cornerstone of visual intelligence, and a key functionality for artificial, autonomous visual machines. This problem will never be solved without algorithms able to adapt and generalize across visual domains. Within the context of domain adaptation and generalization, this paper focuses on the predictive domain adaptation scenario, namely the case where no target data are available and the system has to learn to generalize from annotated source images plus unlabeled samples with associated metadata from auxiliary domains. Our contribution is the first deep architecture that tackles predictive domain adaptation, able to leverage over the information brought by the auxiliary domains through a graph. Moreover, we present a simple yet effective strategy that allows us to take advantage of the incoming target data at test time, in a continuous domain adaptation scenario. Experiments on three benchmark databases support the value of our approach.

We present AdaFrame, a framework that adaptively selects relevant frames on a per-input basis for fast video recognition. AdaFrame contains a Long Short-Term Memory network augmented with a global memory that provides context information for searching which frames to use over time. Trained with policy gradient methods, AdaFrame generates a prediction, determines which frame to observe next, and computes the utility, i.e., expected future rewards, of seeing more frames at each time step. At testing time, AdaFrame exploits predicted utilities to achieve adaptive lookahead inference such that the overall computational costs are reduced without incurring a decrease in accuracy. Extensive experiments are conducted on two large-scale video benchmarks, FCVID and ActivityNet. AdaFrame matches the performance of using all frames with only 8.21 and 8.65 frames on FCVID and ActivityNet, respectively. We further qualitatively demonstrate learned frame usage can indicate the difficulty of making classification decisions; easier samples need fewer frames while harder ones require more, both at instance-level within the same class and at class-level among different categories.

The cosine-based softmax losses and their variants achieve great success in deep learning based face recognition. However, hyperparameter settings in these losses have significant influences on the optimization path as well as the final recognition performance. Manually tuning those hyperparameters heavily relies on user experience and requires many training tricks. In this paper, we investigate in depth the effects of two important hyperparameters of cosine-based softmax losses, the scale parameter and angular margin parameter, by analyzing how they modulate the predicted classification probability. Based on these analysis, we propose a novel cosine-based softmax loss, AdaCos, which is hyperparameter-free and leverages an adaptive scale parameter to automatically strengthen the training supervisions during the training process. We apply the proposed AdaCos loss to large-scale face verification and identification datasets, including LFW, MegaFace, and IJB-C 1:1 Verification. Our results show that training deep neural networks with the AdaCos loss is stable and able to achieve high face recognition accuracy. Our method outperforms state-of-the-art softmax losses on all the three datasets.

Most approaches to visual scene analysis have emphasised parallel processing of the image elements. However, one area in which the sequential nature of vision is apparent, is that of segmenting multiple, potentially similar and partially occluded objects in a scene. In this work, we revisit the recurrent formulation of this challenging problem in the context of reinforcement learning. Motivated by the limitations of the global max-matching assignment of the ground-truth segments to the recurrent states, we develop an actor-critic approach in which the actor recurrently predicts one instance mask at a time and utilises the gradient from a concurrently trained critic network. We formulate the state, action, and the reward such as to let the critic model long-term effects of the current prediction and in- corporate this information into the gradient signal. Furthermore, to enable effective exploration in the inherently high-dimensional action space of instance masks, we learn a compact representation using a conditional variational auto-encoder. We show that our actor-critic model consistently provides accuracy benefits over the recurrent baseline on standard instance segmentation benchmarks.

Weakly supervised object detection aims at reducing the amount of supervision required to train detection models. Such models are traditionally learned from images/videos labelled only with the object class and not the object bounding box. In our work, we try to leverage not only the object class labels but also the action labels associated with the data. We show that the action depicted in the image/video can provide strong cues about the location of the associated object. We learn a spatial prior for the object dependent on the action (e.g. "ball" is closer to "leg of the person" in "kicking ball"), and incorporate this prior to simultaneously train a joint object detection and action classification model. We conducted experiments on both video datasets and image datasets to evaluate the performance of our weakly supervised object detection model. Our approach outperformed the current state-of-the-art (SOTA) method by more than 6% in mAP on the Charades video dataset.

One of the key limitations of traditional machine learning methods is their requirement for training data that exemplifies all the information to be learned. This is a particular problem for visual question answering methods, which may be asked questions about virtually anything. The approach we propose is a step toward overcoming this limitation by searching for the information required at test time. The resulting method dynamically utilizes data from an external source, such as a large set of questions/answers or images/captions. Concretely, we learn a set of base weights for a simple VQA model, that are specifically adapted to a given question with the information specifically retrieved for this question. The adaptation process leverages recent advances in gradient-based meta learning and contributions for efficient retrieval and cross-domain adaptation. We surpass the state-of-the-art on the VQA-CP v2 benchmark and demonstrate our approach to be intrinsically more robust to out-of-distribution test data. We demonstrate the use of external non-VQA data using the MS COCO captioning dataset to support the answering process. This approach opens a new avenue for open-domain VQA systems that interface with diverse sources of data.

Action recognition with skeleton data has recently attracted much attention in computer vision. Previous studies are mostly based on fixed skeleton graphs, only capturing local physical dependencies among joints, which may miss implicit joint correlations. To capture richer dependencies, we introduce an encoder-decoder structure, called A-link inference module, to capture action-specific latent dependencies, i.e. actional links, directly from actions. We also extend the existing skeleton graphs to represent higher-order dependencies, i.e. structural links. Combing the two types of links into a generalized skeleton graph, We further propose the actional-structural graph convolution network (AS-GCN), which stacks actional-structural graph convolution and temporal convolution as a basic building block, to learn both spatial and temporal features for action recognition. A future pose prediction head is added in parallel to the recognition head to help capture more detailed action patterns through self-supervision. We validate AS-GCN in action recognition using two skeleton data sets, NTU-RGB+D and Kinetics. The proposed AS-GCN achieves consistently large improvement compared to the state-of-the-art methods. As a side product, AS-GCN also shows promising results for future pose prediction.

Recognizing every person's action in a crowded and cluttered environment is a challenging task in computer vision. We propose to tackle this challenging problem using a holistic 4D "scan" of a cluttered scene to include every detail about the people and environment. This leads to a new problem, i.e., recognizing multiple people's actions in the cluttered 4D representation. At the first step, we propose a new method to track people in 4D, which can reliably detect and follow each person in real time. Then, we build a new deep neural network, the Action4DNet, to recognize the action of each tracked person. Such a model gives reliable and accurate results in the real-world settings. We also design an adaptive 3D convolution layer and a novel discriminative temporal feature learning objective to further improve the performance of our model. Our method is invariant to camera view angles, resistant to clutter and able to handle crowd. The experimental results show that the proposed method is fast, reliable and accurate. Our method paves the way to action recognition in the real-world applications and is ready to be deployed to enable smart homes, smart factories and smart stores.

Recognising actions in videos relies on labelled supervision during training, typically the start and end times of each action instance. This supervision is not only subjective, but also expensive to acquire. Weak video-level supervision has been successfully exploited for recognition in untrimmed videos, however it is challenged when the number of different actions in training videos increases. We propose a method that is supervised by single timestamps located around each action instance, in untrimmed videos. We replace expensive action bounds with sampling distributions initialised from these timestamps. We then use the classifier's response to iteratively update the sampling distributions. We demonstrate that these distributions converge to the location and extent of discriminative action segments. We evaluate our method on three datasets for fine-grained recognition, with increasing number of different actions per video, and show that single timestamps offer a reasonable compromise between recognition performance and labelling effort, performing comparably to full temporal supervision. Our update method improves top-1 test accuracy by up to 5.4%. across the evaluated datasets.

Non-line-of-sight (NLOS) imaging enables unprecedented capabilities in a wide range of applications, including robotic and machine vision, remote sensing, autonomous vehicle navigation, and medical imaging. Recent approaches to solving this challenging problem employ optical time-of-flight imaging systems with highly sensitive time-resolved photodetectors and ultra-fast pulsed lasers. However, despite recent successes in NLOS imaging using these systems, widespread implementation and adoption of the technology remains a challenge because of the requirement for specialized, expensive hardware. We introduce acoustic NLOS imaging, which is orders of magnitude less expensive than most optical systems and captures hidden 3D geometry at longer ranges with shorter acquisition times compared to state-of-the-art optical methods. Inspired by hardware setups used in radar and algorithmic approaches to model and invert wave-based image formation models developed in the seismic imaging community, we demonstrate a new approach to seeing around corners.

The deep learning revolution brought us an extensive array of neural network architectures that achieve state-of-the-art performance in a wide variety of Computer Vision tasks including among others, classification, detection and segmentation. In parallel, we have also been observing an unprecedented demand in computational and memory requirements, rendering the efficient use of neural networks in low-powered devices virtually unattainable. Towards this end, we propose a three-stage compression and acceleration pipeline that sparsifies, quantizes and entropy encodes activation maps of Convolutional Neural Networks. Sparsification increases the representational power of activation maps leading to both acceleration of inference and higher model accuracy. Inception-V3 and MobileNet-V1 can be accelerated by as much as 1.6x with an increase in accuracy of 0.38% and 0.54% on the ImageNet and CIFAR-10 datasets respectively. Quantizing and entropy coding the sparser activation maps lead to higher compression over the baseline, reducing the memory cost of the network execution. Inception-V3 and MobileNet-V1 activation maps, quantized to 16 bits, are compressed by as much as 6x with an increase in accuracy of 0.36% and 0.55% respectively.

We present Accel, a novel semantic video segmentation system that achieves high accuracy at low inference cost by combining the predictions of two network branches: (1) a reference branch that extracts high-detail features on a reference keyframe, and warps these features forward using frame-to-frame optical flow estimates, and (2) an update branch that computes features of adjustable quality on the current frame, performing a temporal update at each video frame. The modularity of the update branch, where feature subnetworks of varying layer depth can be inserted (e.g. ResNet-18 to ResNet-101), enables operation over a new, state-of-the-art accuracy-throughput trade-off spectrum. Over this curve, Accel models achieve both higher accuracy and faster inference times than the closest comparable single-frame segmentation networks. In general, Accel significantly outperforms previous work on efficient semantic video segmentation, correcting warping-related error that compounds on datasets with complex dynamics. Accel is end-to-end trainable and highly modular: the reference network, the optical flow network, and the update network can each be selected independently, depending on application requirements, and then jointly fine-tuned. The result is a robust, general system for fast, high-accuracy semantic segmentation on video.

While recent years have witnessed astonishing improvements in visual tracking robustness, the advancements in tracking accuracy have been limited. As the focus has been directed towards the development of powerful classifiers, the problem of accurate target state estimation has been largely overlooked. In fact, most trackers resort to a simple multi-scale search in order to estimate the target bounding box. We argue that this approach is fundamentally limited since target estimation is a complex task, requiring high-level knowledge about the object. We address this problem by proposing a novel tracking architecture, consisting of dedicated target estimation and classification components. High level knowledge is incorporated into the target estimation through extensive offline learning. Our target estimation component is trained to predict the overlap between the target object and an estimated bounding box. By carefully integrating target-specific information, our approach achieves previously unseen bounding box accuracy. We further introduce a classification component that is trained online to guarantee high discriminative power in the presence of distractors. Our final tracking framework sets a new state-of-the-art on five challenging benchmarks. On the new large-scale TrackingNet dataset, our tracker ATOM achieves a relative gain of 15% over the previous best approach, while running at over 30 FPS. Code and models are available at https://github.com/visionml/pytracking.

Significant progress has been made with image stylization using deep learning, especially with generative adversarial networks (GANs). However, existing methods fail to produce high quality artistic portrait drawings. Such drawings have a highly abstract style, containing a sparse set of continuous graphical elements such as lines, and so small artifacts are much more exposed than for painting styles. Moreover, artists tend to use different strategies to draw different facial features and the lines drawn are only loosely related to obvious image features. To address these challenges, we propose APDrawingGAN, a novel GAN based architecture that builds upon hierarchical generators and discriminators combining both a global network (for images as a whole) and local networks (for individual facial regions). This allows dedicated drawing strategies to be learned for different facial features. Since artists' drawings may not have lines perfectly aligned with image features, we develop a novel loss to measure similarity between generated and artists' drawings based on distance transforms, leading to improved strokes in portrait drawing. To train APDrawingGAN, we construct an artistic drawing dataset containing high-resolution portrait photos and corresponding professional artistic drawings. Extensive experiments, including a user study, show that APDrawingGAN produces significantly better artistic drawings than state-of-the-art methods.

Neural architectures are the foundation for improving performance of deep neural networks (DNNs). This paper presents deep compositional grammatical architectures which harness the best of two worlds: grammar models and DNNs. The proposed architectures integrate compositionality and reconfigurability of the former and the capability of learning rich features of the latter in a principled way. We utilize AND-OR Grammar (AOG) as network generator in this paper and call the resulting networks AOGNets. An AOGNet consists of a number of stages each of which is composed of a number of AOG building blocks. An AOG building block splits its input feature map into N groups along feature channels and then treat it as a sentence of N words. It then jointly realizes a phrase structure grammar and a dependency grammar in bottom-up parsing the "sentence" for better feature exploration and reuse. It provides a unified framework for the best practices developed in state-of-the-art DNNs. In experiments, AOGNet is tested in the ImageNet-1K classification benchmark and the MS-COCO object detection and segmentation benchmark. In ImageNet-1K, AOGNet obtains better performance than ResNet and most of its variants, ResNeXt and its attention based variants such as SENet, DenseNet and DualPathNet. AOGNet also obtains the best model interpretability score using network dissection. AOGNet further shows better potential in adversarial defense. In MS-COCO, AOGNet obtains better performance than the ResNet and ResNeXt backbones in Mask R-CNN.

Image repurposing is a commonly used method for spreading misinformation on social media and online forums, which involves publishing untampered images with modified metadata to create rumors and further propaganda. While manual verification is possible, given vast amounts of verified knowledge available on the internet, the increasing prevalence and ease of this form of semantic manipulation call for the development of robust automatic ways of assessing the semantic integrity of multimedia data. In this paper, we present a novel method for image repurposing detection that is based on the real-world adversarial interplay between a bad actor who repurposes images with counterfeit metadata and a watchdog who verifies the semantic consistency between images and their accompanying metadata, where both players have access to a reference dataset of verified content, which they can use to achieve their goals. The proposed method exhibits state-of-the-art performance on location-identity, subject-identity and painting-artist verification, showing its efficacy across a diverse set of scenarios.

The success of deep neural networks often relies on a large amount of labeled examples, which can be difficult to obtain in many real scenarios. To address this challenge, unsupervised methods are strongly preferred for training neural networks without using any labeled data. In this paper, we present a novel paradigm of unsupervised representation learning by Auto-Encoding Transformation (AET) in contrast to the conventional Auto-Encoding Data (AED) approach. Given a randomly sampled transformation, AET seeks to predict it merely from the encoded features as accurately as possible at the output end. The idea is the following: as long as the unsupervised features successfully encode the essential information about the visual structures of original and transformed images, the transformation can be well predicted. We will show that this AET paradigm allows us to instantiate a large variety of transformations, from parameterized, to non-parameterized and GAN-induced ones. Our experiments show that AET greatly improves over existing unsupervised approaches, setting new state-of-the-art performances being greatly closer to the upper bounds by their fully supervised counterparts on CIFAR-10, ImageNet and Places datasets.

Learning on data represented with multiple views (e.g., multiple types of descriptors or modalities) is a rapidly growing direction in machine learning and computer vision. Although effectiveness achieved, most existing algorithms usually focus on classification or clustering tasks. Differently, in this paper, we focus on unsupervised representation learning and propose a novel framework termed Autoencoder in Autoencoder Networks (AE^2-Nets), which integrates information from heterogeneous sources into an intact representation by the nested autoencoder framework. The proposed method has the following merits: (1) our model jointly performs view-specific representation learning (with the inner autoencoder networks) and multi-view information encoding (with the outer autoencoder networks) in a unified framework; (2) due to the degradation process from the latent representation to each single view, our model flexibly balances the complementarity and consistence among multiple views. The proposed model is efficiently solved by the alternating direction method (ADM), and demonstrates the effectiveness compared with state-of-the-art algorithms.