ICCV 2019 Papers

We present a new data-driven method for robust skin detection from a single human portrait image. Unlike previous methods, we incorporate human body as a weak semantic guidance into this task, considering acquiring large-scale of human labeled skin data is commonly expensive and time-consuming. To be specific, we propose a dual-task neural network for joint detection of skin and body via a semi-supervised learning strategy. The dual-task network contains a shared encoder but two decoders for skin and body separately. For each decoder, its output also serves as a guidance for its counterpart, making both decoders mutually guided. Extensive experiments were conducted to demonstrate the effectiveness of our network with mutual guidance, and experimental results show our network outperforms the state-of-the-art in skin detection.

We propose a GAN-based scene-specific instance synthesis and classification model for semi-supervised pedestrian detection. Instead of collecting unreliable detections from unlabeled data, we adopt a class-conditional GAN for synthesizing pedestrian instances to alleviate the problem of insufficient labeled data. With the help of a base detector, we integrate pedestrian instance synthesis and detection by including a post-refinement classifier (PRC) into a minimax game. A generator and the PRC can mutually reinforce each other by synthesizing high-fidelity pedestrian instances and providing more accurate categorical information. Both of them compete with a class-conditional discriminator and a class-specific discriminator, such that the four fundamental networks in our model can be jointly trained. In our experiments, we validate that the proposed model significantly improves the performance of the base detector and achieves state-of-the-art results on multiple benchmarks. As shown in Figure 1, the result indicates the possibility of using inexpensively synthesized instances for improving semi-supervised detection models.

Monocular face reconstruction is a challenging task in computer vision, which aims to recover 3D face geometry from a single RGB face image. Recently, deep learning based methods have achieved great improvements on monocular face reconstruction. However, for deep learning-based methods to reach optimal performance, it is paramount to have large-scale training images with ground-truth 3D face geometry, which is generally difficult for human to annotate. To tackle this problem, we propose a semi-supervised monocular reconstruction method, which jointly optimizes a shape-preserved domain-transfer CycleGAN and a shape estimation network. The framework is semi-supervised trained with 3D rendered images with ground-truth shapes and in-the-wild face images without any extra annotation. The CycleGAN network transforms all realistic images to have the rendered style and is end-to-end trained within the overall framework. This is the key difference compared with existing CycleGAN-based learning methods, which just used CycleGAN as a separate training sample generator. Novel landmark consistency loss and edge-aware shape estimation loss are proposed for our two networks to jointly solve the challenging face reconstruction problem. Extensive experiments on public face reconstruction datasets demonstrate the effectiveness of our overall method as well as the individual components.

In this work, we propose a simple yet effective semi-supervised learning approach called Augmented Distribution Alignment. We reveal that an essential sampling bias exists in semi-supervised learning due to the limited number of labeled samples, which often leads to a considerable empirical distribution mismatch between labeled data and unlabeled data. To this end, we propose to align the empirical distributions of labeled and unlabeled data to alleviate the bias. On one hand, we adopt an adversarial training strategy to minimize the distribution distance between labeled and unlabeled data as inspired by domain adaptation works. On the other hand, to deal with the small sample size issue of labeled data, we also propose a simple interpolation strategy to generate pseudo training samples. Those two strategies can be easily implemented into existing deep neural networks. We demonstrate the effectiveness of our proposed approach on the benchmark SVHN and CIFAR10 datasets. Our code is available at https://github.com/qinenergy/adanet .

Contemporary domain adaptation methods are very effective at aligning feature distributions of source and target domains without any target supervision. However, we show that these techniques perform poorly when even a few labeled examples are available in the target domain. To address this semi-supervised domain adaptation (SSDA) setting, we propose a novel Minimax Entropy (MME) approach that adversarially optimizes an adaptive few-shot model. Our base model consists of a feature encoding network, followed by a classification layer that computes the features' similarity to estimated prototypes (representatives of each class). Adaptation is achieved by alternately maximizing the conditional entropy of unlabeled target data with respect to the classifier and minimizing it with respect to the feature encoder. We empirically demonstrate the superiority of our method over many baselines, including conventional feature alignment and few-shot methods, setting a new state of the art for SSDA. Our code is available at http://cs-people.bu.edu/keisaito/research/MME.html.

This paper presents a new model, Semantics-enhanced Generative Adversarial Network (SEGAN), for fine-grained text-to-image generation. We introduce two modules, a Semantic Consistency Module (SCM) and an Attention Competition Module (ACM), to our SEGAN. The SCM incorporates image-level semantic consistency into the training of the Generative Adversarial Network (GAN), and can diversify the generated images and improve their structural coherence. A Siamese network and two types of semantic similarities are designed to map the synthesized image and the groundtruth image to nearby points in the latent semantic feature space. The ACM constructs adaptive attention weights to differentiate keywords from unimportant words, and improves the stability and accuracy of SEGAN. Extensive experiments demonstrate that our SEGAN significantly outperforms existing state-of-the-art methods in generating photo-realistic images. All source codes and models will be released for comparative study.

Semantic scene understanding is important for various applications. In particular, self-driving cars need a fine-grained understanding of the surfaces and objects in their vicinity. Light detection and ranging (LiDAR) provides precise geometric information about the environment and is thus a part of the sensor suites of almost all self-driving cars. Despite the relevance of semantic scene understanding for this application, there is a lack of a large dataset for this task which is based on an automotive LiDAR. In this paper, we introduce a large dataset to propel research on laser-based semantic segmentation. We annotated all sequences of the KITTI Vision Odometry Benchmark and provide dense point-wise annotations for the complete 360-degree field-of-view of the employed automotive LiDAR. We propose three benchmark tasks based on this dataset: (i) semantic segmentation of point clouds using a single scan, (ii) semantic segmentation using multiple past scans, and (iii) semantic scene completion, which requires to anticipate the semantic scene in the future. We provide baseline experiments and show that there is a need for more sophisticated models to efficiently tackle these tasks. Our dataset opens the door for the development of more advanced methods, but also provides plentiful data to investigate new research directions.

Weakly-supervised learning under image-level labels supervision has been widely applied to semantic segmentation of medical lesions regions. However, 1) most existing models rely on effective constraints to explore the internal representation of lesions, which only produces inaccurate and coarse lesions regions; 2) they ignore the strong probabilistic dependencies between target lesions dataset (e.g., enteroscopy images) and well-to-annotated source diseases dataset (e.g., gastroscope images). To better utilize these dependencies, we present a new semantic lesions representation transfer model for weakly-supervised endoscopic lesions segmentation, which can exploit useful knowledge from relevant fully-labeled diseases segmentation task to enhance the performance of target weakly-labeled lesions segmentation task. More specifically, a pseudo label generator is proposed to leverage seed information to generate highly-confident pseudo pixel labels by incorporating class balance and super-pixel spatial prior. It can iteratively include more hard-to-transfer samples from weakly-labeled target dataset into training set. Afterwards, dynamically-searched feature centroids for same class among different datasets are aligned by accumulating previously-learned features. Meanwhile, adversarial learning is also employed in this paper, to narrow the gap between the lesions among different datasets in output space. Finally, we build a new medical endoscopic dataset with 3659 images collected from more than 1100 volunteers. Extensive experiments on our collected dataset and several benchmark datasets validate the effectiveness of our model.

Sketch-based image retrieval (SBIR) is widely recognized as an important vision problem which implies a wide range of real-world applications. Recently, research interests arise in solving this problem under the more realistic and challenging setting of zero-shot learning. In this paper, we investigate this problem from the viewpoint of domain adaptation which we show is critical in improving feature embedding in the zero-shot scenario. Based on a framework which starts with a pre-trained model on ImageNet and fine-tunes it on the training set of SBIR benchmark, we advocate the importance of preserving previously acquired knowledge, e.g., the rich discriminative features learned from ImageNet, to improve the model's transfer ability. For this purpose, we design an approach named Semantic-Aware Knowledge prEservation (SAKE), which fine-tunes the pre-trained model in an economical way and leverages semantic information, e.g., inter-class relationship, to achieve the goal of knowledge preservation. Zero-shot experiments on two extended SBIR datasets, TU-Berlin and Sketchy, verify the superior performance of our approach. Extensive diagnostic experiments validate that knowledge preserved benefits SBIR in zero-shot settings, as a large fraction of the performance gain is from the more properly structured feature embedding for photo images.

The accuracy of stereo matching has been greatly improved by using deep learning with convolutional neural networks. To further capture the details of disparity maps, in this paper, we propose a novel semantic stereo network named SSPCV-Net, which includes newly designed pyramid cost volumes for describing semantic and spatial information on multiple levels. The semantic features are inferred by a semantic segmentation subnetwork while the spatial features are derived by hierarchical spatial pooling. In the end, we design a 3D multi-cost aggregation module to integrate the extracted multilevel features and perform regression for accurate disparity maps. We conduct comprehensive experiments and comparisons with some recent stereo matching networks on Scene Flow, KITTI 2015 and 2012, and Cityscapes benchmark datasets, and the results show that the proposed SSPCV-Net significantly promotes the state-of-the-art stereo-matching performance.

Detecting semantic parts of an object is a challenging task, particularly because it is hard to annotate semantic parts and construct large datasets. In this paper, we present an approach which can learn from a small annotated dataset containing a limited range of viewpoints and generalize to detect semantic parts for a much larger range of viewpoints. The approach is based on our matching algorithm, which is used for finding accurate spatial correspondence between two images and transplanting semantic parts annotated on one image to the other. Images in the training set are matched to synthetic images rendered from a 3D CAD model, following which a clustering algorithm is used to automatically annotate semantic parts of the CAD model. During the testing period, this CAD model can synthesize annotated images under every viewpoint. These synthesized images are matched to images in the testing set to detect semantic parts in novel viewpoints. Our algorithm is simple, intuitive, and contains very few parameters. Experiments show our method outperforms standard deep learning approaches and, in particular, performs much better on novel viewpoints. For facilitating the future research, code is available: https://github.com/ytongbai/SemanticPartDetection

Deep neural networks have been shown to exhibit an intriguing vulnerability to adversarial input images corrupted with imperceptible perturbations. However, the majority of adversarial attacks assume global, fine-grained control over the image pixel space. In this paper, we consider a different setting: what happens if the adversary could only alter specific attributes of the input image? These would generate inputs that might be perceptibly different, but still natural-looking and enough to fool a classifier. We propose a novel approach to generate such "semantic" adversarial examples by optimizing a particular adversarial loss over the range-space of a parametric conditional generative model. We demonstrate implementations of our attacks on binary classifiers trained on face images, and show that such natural-looking semantic adversarial examples exist. We evaluate the effectiveness of our attack on synthetic and real data, and present detailed comparisons with existing attack methods. We supplement our empirical results with theoretical bounds that demonstrate the existence of such parametric adversarial examples.

Deep learning-based object detectors have shown remarkable improvements. However, supervised learning-based methods perform poorly when the train data and the test data have different distributions. To address the issue, domain adaptation transfers knowledge from the label-sufficient domain (source domain) to the label-scarce domain (target domain). Self-training is one of the powerful ways to achieve domain adaptation since it helps class-wise domain adaptation. Unfortunately, a naive approach that utilizes pseudo-labels as ground-truth degenerates the performance due to incorrect pseudo-labels. In this paper, we introduce a weak self-training (WST) method and adversarial background score regularization (BSR) for domain adaptive one-stage object detection. WST diminishes the adverse effects of inaccurate pseudo-labels to stabilize the learning procedure. BSR helps the network extract discriminative features for target backgrounds to reduce the domain shift. Two components are complementary to each other as BSR enhances discrimination between foregrounds and backgrounds, whereas WST strengthen class-wise discrimination. Experimental results show that our approach effectively improves the performance of the one-stage object detection in unsupervised domain adaptation setting.

Person re-identification (Re-ID) has achieved great improvement with deep learning and a large amount of labelled training data. However, it remains a challenging task for adapting a model trained in a source domain of labelled data to a target domain of only unlabelled data available. In this work, we develop a self-training method with progressive augmentation framework (PAST) to promote the model performance progressively on the target dataset. Specially, our PAST framework consists of two stages, namely, conservative stage and promoting stage. The conservative stage captures the local structure of target-domain data points with triplet-based loss functions, leading to improved feature representations. The promoting stage continuously optimizes the network by appending a changeable classification layer to the last layer of the model, enabling the use of global information about the data distribution. Importantly, we propose a new self-training strategy that progressively augments the model capability by adopting conservative and promoting stages alternately. Furthermore, to improve the reliability of selected triplet samples, we introduce a ranking-based triplet loss in the conservative stage, which is a label-free objective function based on the similarities between data pairs. Experiments demonstrate that the proposed method achieves state-of-the-art person Re-ID performance under the unsupervised cross-domain setting. Code is available at: tinyurl.com/PASTReID

In this paper, we propose a novel self-supervised representation learning by taking advantage of a neighborhood-relational encoding (NRE) among the training data. Conventional unsupervised learning methods only focused on training deep networks to understand the primitive characteristics of the visual data, mainly to be able to reconstruct the data from a latent space. They often neglected the relation among the samples, which can serve as an important metric for self-supervision. Different from the previous work, NRE aims at preserving the local neighborhood structure on the data manifold. Therefore, it is less sensitive to outliers. We integrate our NRE component with an encoder-decoder structure for learning to represent samples considering their local neighborhood information. Such discriminative and unsupervised representation learning scheme is adaptable to different computer vision tasks due to its independence from intense annotation requirements. We evaluate our proposed method for different tasks, including classification, detection, and segmentation based on the learned latent representations. In addition, we adopt the auto-encoding capability of our proposed method for applications like defense against adversarial example attacks and video anomaly detection. Results confirm the performance of our method is better or at least comparable with the state-of-the-art for each specific application, but with a generic and self-supervised approach.

We present an information-theoretically motivated constraint for self-supervised representation learning from multiple related domains. In contrast to previous self-supervised learning methods, our approach learns from multiple domains, which has the benefit of decreasing the build-in bias of individual domain, as well as leveraging information and allowing knowledge transfer across multiple domains. The proposed mutual information constraints encourage neural network to extract common invariant information across domains and to preserve peculiar information of each domain simultaneously. We adopt tractable upper and lower bounds of mutual information to make the proposed constraints solvable. The learned representation is more unbiased and robust toward the input images. Extensive experimental results on both multi-domain and large-scale datasets demonstrate the necessity and advantage of multi-domain self-supervised learning with mutual information constraints. Representations learned in our framework on state-of-the-art methods achieve improved performance than those learned on a single domain.

Humans are able to localize objects in the environment using both visual and auditory cues, integrating information from multiple modalities into a common reference frame. We introduce a system that can leverage unlabeled audiovisual data to learn to localize objects (moving vehicles) in a visual reference frame, purely using stereo sound at inference time. Since it is labor-intensive to manually annotate the correspondences between audio and object bounding boxes, we achieve this goal by using the co-occurrence of visual and audio streams in unlabeled videos as a form of self-supervision, without resorting to the collection of ground truth annotations. In particular, we propose a framework that consists of a vision "teacher" network and a stereo-sound "student" network. During training, knowledge embodied in a well-established visual vehicle detection model is transferred to the audio domain using unlabeled videos as a bridge. At test time, the stereo-sound student network can work independently to perform object localization using just stereo audio and camera meta-data, without any visual input. Experimental results on a newly collected Auditory Vehicles Tracking dataset verify that our proposed approach outperforms several baseline approaches. We also demonstrate that our cross-modal auditory localization approach can assist in the visual localization of moving vehicles under poor lighting conditions.

Monocular depth estimators can be trained with various forms of self-supervision from binocular-stereo data to circumvent the need for high-quality laser-scans or other ground-truth data. The disadvantage, however, is that the photometric reprojection losses used with self-supervised learning typically have multiple local minima. These plausible-looking alternatives to ground-truth can restrict what a regression network learns, causing it to predict depth maps of limited quality. As one prominent example, depth discontinuities around thin structures are often incorrectly estimated by current state-of-the-art methods. Here, we study the problem of ambiguous reprojections in depth-prediction from stereo-based self-supervision, and introduce Depth Hints to alleviate their effects. Depth Hints are complementary depth-suggestions obtained from simple off-the-shelf stereo algorithms. These hints enhance an existing photometric loss function, and are used to guide a network to learn better weights. They require no additional data, and are assumed to be right only sometimes. We show that using our Depth Hints gives a substantial boost when training several leading self-supervised-from-stereo models, not just our own. Further, combined with other good practices, we produce state-of-the-art depth predictions on the KITTI benchmark.

We present GLNet, a self-supervised framework for learning depth, optical flow, camera pose and intrinsic parameters from monocular video -- addressing the difficulty of acquiring realistic ground-truth for such tasks. We propose three contributions: 1) we design new loss functions that capture multiple geometric constraints (eg. epipolar geometry) as well as adaptive photometric loss that supports multiple moving objects, rigid and non-rigid, 2) we extend the model such that it predicts camera intrinsics, making it applicable to uncalibrated video, and 3) we propose several online refinement strategies that rely on the symmetry of our self-supervised loss in training and testing, in particular optimizing model parameters and/or the output of different tasks, leveraging their mutual interactions. The idea of jointly optimizing the system output, under all geometric and photometric constraints can be viewed as a dense generalization of classical bundle adjustment. We demonstrate the effectiveness of our method on KITTI and Cityscapes, where we outperform previous self-supervised approaches on multiple tasks. We also show good generalization for transfer learning.