ICCV 2019 论文列表

Depth perception is considered an invaluable source of information for various vision tasks. However, depth maps acquired using consumer-level sensors still suffer from non-negligible noise. This fact has recently motivated researchers to exploit traditional filters, as well as the deep learning paradigm, in order to suppress the aforementioned non-uniform noise, while preserving geometric details. Despite the effort, deep depth denoising is still an open challenge mainly due to the lack of clean data that could be used as ground truth. In this paper, we propose a fully convolutional deep autoencoder that learns to denoise depth maps, surpassing the lack of ground truth data. Specifically, the proposed autoencoder exploits multiple views of the same scene from different points of view in order to learn to suppress noise in a self-supervised end-to-end manner using depth and color information during training, yet only depth during inference. To enforce self-supervision, we leverage a differentiable rendering technique to exploit photometric supervision, which is further regularized using geometric and surface priors. As the proposed approach relies on raw data acquisition, a large RGB-D corpus is collected using Intel RealSense sensors. Complementary to a quantitative evaluation, we demonstrate the effectiveness of the proposed self-supervised denoising approach on established 3D reconstruction applications. Code is avalable at https://github.com/VCL3D/DeepDepthDenoising

Domain adaptation in person re-identification (re-ID) has always been a challenging task. In this work, we explore how to harness the similar natural characteristics existing in the samples from the target domain for learning to conduct person re-ID in an unsupervised manner. Concretely, we propose a Self-similarity Grouping (SSG) approach, which exploits the potential similarity (from the global body to local parts) of unlabeled samples to build multiple clusters from different views automatically. These independent clusters are then assigned with labels, which serve as the pseudo identities to supervise the training process. We repeatedly and alternatively conduct such a grouping and training process until the model is stable. Despite the apparent simplify, our SSG outperforms the state-of-the-arts by more than 4.6% (DukeMTMC-Market1501) and 4.4% (Market1501-DukeMTMC) in mAP, respectively. Upon our SSG, we further introduce a clustering-guided semisupervised approach named SSG ++ to conduct the one-shot domain adaption in an open set setting (i.e. the number of independent identities from the target domain is unknown). Without spending much effort on labeling, our SSG ++ can further promote the mAP upon SSG by 10.7% and 6.9%, respectively. Our Code is available at: https://github.com/OasisYang/SSG .

During the past years, tremendous advances in image restoration tasks have been achieved using highly complex neural networks. Despite their good restoration performance, the heavy computational burden hinders the deployment of these networks on constrained devices, e.g. smart phones and consumer electronic products. To tackle this problem, we propose a self-guided network (SGN), which adopts a top-down self-guidance architecture to better exploit image multi-scale information. SGN directly generates multi-resolution inputs with the shuffling operation. Large-scale contextual information extracted at low resolution is gradually propagated into the higher resolution sub-networks to guide the feature extraction processes at these scales. Such a self-guidance strategy enables SGN to efficiently incorporate multi-scale information and extract good local features to recover noisy images. We validate the effectiveness of SGN through extensive experiments. The experimental results demonstrate that SGN greatly improves the memory and runtime efficiency over state-of-the-art efficient methods, without trading off PSNR accuracy.

Deep learning-based semantic segmentation methods have an intrinsic limitation that training a model requires a large amount of data with pixel-level annotations. To address this challenging issue, many researchers give attention to unsupervised domain adaptation for semantic segmentation. Unsupervised domain adaptation seeks to adapt the model trained on the source domain to the target domain. In this paper, we introduce a self-ensembling technique, one of the successful methods for domain adaptation in classification. However, applying self-ensembling to semantic segmentation is very difficult because heavily-tuned manual data augmentation used in self-ensembling is not useful to reduce the large domain gap in the semantic segmentation. To overcome this limitation, we propose a novel framework consisting of two components, which are complementary to each other. First, we present a data augmentation method based on Generative Adversarial Networks (GANs), which is computationally efficient and effective to facilitate domain alignment. Given those augmented images, we apply self-ensembling to enhance the performance of the segmentation network on the target domain. The proposed method outperforms state-of-the-art semantic segmentation methods on unsupervised domain adaptation benchmarks.

In this paper, we propose a self-critical attention learning method for person re-identification. Unlike most existing methods which train the attention mechanism in a weakly-supervised manner and ignore the attention confidence level, we learn the attention with a critic which measures the attention quality and provides a powerful supervisory signal to guide the learning process. Moreover, the critic model facilitates the interpretation of the effectiveness of the attention mechanism during the learning process, by estimating the quality of the attention maps. Specifically, we jointly train our attention agent and critic in a reinforcement learning manner, where the agent produces the visual attention while the critic analyzes the gain from the attention and guides the agent to maximize this gain. We design spatial- and channel-wise attention models with our critic module and evaluate them on three popular benchmarks including Market-1501, DukeMTMC-ReID, and CUHK03. The experimental results demonstrate the superiority of our method, which outperforms the state-of-the-art methods by a large margin of 5.9%/2.1%, 6.3%/3.0%, and 10.5%/9.5% on mAP/Rank-1, respectively.

Typically, a salient object detection (SOD) model faces opposite requirements in processing object interiors and boundaries. The features of interiors should be invariant to strong appearance change so as to pop-out the salient object as a whole, while the features of boundaries should be selective to slight appearance change to distinguish salient objects and background. To address this selectivity-invariance dilemma, we propose a novel boundary-aware network with successive dilation for image-based SOD. In this network, the feature selectivity at boundaries is enhanced by incorporating a boundary localization stream, while the feature invariance at interiors is guaranteed with a complex interior perception stream. Moreover, a transition compensation stream is adopted to amend the probable failures in transitional regions between interiors and boundaries. In particular, an integrated successive dilation module is proposed to enhance the feature invariance at interiors and transitional regions. Extensive experiments on six datasets show that the proposed approach outperforms 16 state-of-the-art methods.

Fine-grained recognition poses the unique challenge of capturing subtle inter-class differences under considerable intra-class variances (e.g., beaks for bird species). Conventional approaches crop local regions and learn detailed representation from those regions, but suffer from the fixed number of parts and missing of surrounding context. In this paper, we propose a simple yet effective framework, called Selective Sparse Sampling, to capture diverse and fine-grained details. The framework is implemented using Convolutional Neural Networks, referred to as Selective Sparse Sampling Networks (S3Ns). With image-level supervision, S3Ns collect peaks, i.e., local maximums, from class response maps to estimate informative, receptive fields and learn a set of sparse attention for capturing fine-detailed visual evidence as well as preserving context. The evidence is selectively sampled to extract discriminative and complementary features, which significantly enrich the learned representation and guide the network to discover more subtle cues. Extensive experiments and ablation studies show that the proposed method consistently outperforms the state-of-the-art methods on challenging benchmarks including CUB-200-2011, FGVC-Aircraft, and Stanford Cars.

Almost all existing deep learning approaches for semantic segmentation tackle this task as a pixel-wise classification problem. Yet humans understand a scene not in terms of pixels, but by decomposing it into perceptual groups and structures that are the basic building blocks of recognition. This motivates us to propose an end-to-end pixel-wise metric learning approach that mimics this process. In our approach, the optimal visual representation determines the right segmentation within individual images and associates segments with the same semantic classes across images. The core visual learning problem is therefore to maximize the similarity within segments and minimize the similarity between segments. Given a model trained this way, inference is performed consistently by extracting pixel-wise embeddings and clustering, with the semantic label determined by the majority vote of its nearest neighbors from an annotated set. As a result, we present the SegSort, as a first attempt using deep learning for unsupervised semantic segmentation, achieving 76% performance of its supervised counterpart. When supervision is available, SegSort shows consistent improvements over conventional approaches based on pixel-wise softmax training. Additionally, our approach produces more precise boundaries and consistent region predictions. The proposed SegSort further produces an interpretable result, as each choice of label can be easily understood from the retrieved nearest segments.

Embodied Question Answering (EQA) is a newly defined research area where an agent is required to answer the user's questions by exploring the real world environment. It has attracted increasing research interests due to its broad applications in automatic driving system, in-home robots, and personal assistants. Most of the existing methods perform poorly in terms of answering and navigation accuracy due to the absence of local details and vulnerability to the ambiguity caused by complicated vision conditions. To tackle these problems, we propose a segmentation based visual attention mechanism for Embodied Question Answering. Firstly, We extract the local semantic features by introducing a novel high-speed video segmentation framework. Then by the guide of extracted semantic features, a bottom-up visual attention mechanism is proposed for the Visual Question Answering (VQA) sub-task. Further, a feature fusion strategy is proposed to guide the training of the navigator without much additional computational cost. The ablation experiments show that our method boosts the performance of VQA module by 4.2% (68.99% vs 64.73%) and leads to 3.6% (48.59% vs 44.98%) overall improvement in EQA accuracy.

Despite the success of Generative Adversarial Networks (GANs), mode collapse remains a serious issue during GAN training. To date, little work has focused on understanding and quantifying which modes have been dropped by a model. In this work, we visualize mode collapse at both the distribution level and the instance level. First, we deploy a semantic segmentation network to compare the distribution of segmented objects in the generated images with the target distribution in the training set. Differences in statistics reveal object classes that are omitted by a GAN. Second, given the identified omitted object classes, we visualize the GAN's omissions directly. In particular, we compare specific differences between individual photos and their approximate inversions by a GAN. To this end, we relax the problem of inversion and solve the tractable problem of inverting a GAN layer instead of the entire generator. Finally, we use this framework to analyze several recent GANs trained on multiple datasets and identify their typical failure cases.

Deep learning has recently been applied with impressive results to extreme low-light imaging. Despite the success of single-image processing, extreme low-light video processing is still intractable due to the difficulty of collecting raw video data with corresponding ground truth. Collecting long-exposure ground truth, as was done for single-image processing, is not feasible for dynamic scenes. In this paper, we present deep processing of very dark raw videos: on the order of one lux of illuminance. To support this line of work, we collect a new dataset of raw low-light videos, in which high-resolution raw data is captured at video rate. At this level of darkness, the signal-to-noise ratio is extremely low (negative if measured in dB) and the traditional image processing pipeline generally breaks down. A new method is presented to address this challenging problem. By carefully designing a learning-based pipeline and introducing a new loss function to encourage temporal stability, we train a siamese network on static raw videos, for which ground truth is available, such that the network generalizes to videos of dynamic scenes at test time. Experimental results demonstrate that the presented approach outperforms state-of-the-art models for burst processing, per-frame processing, and blind temporal consistency.

Motivated by the conventional grouping techniques to image segmentation, we develop their DNN counterpart to tackle the referring variant. The proposed method is driven by a convolutional-recurrent neural network (ConvRNN) that iteratively carries out top-down processing of bottom-up segmentation cues. Given a natural language referring expression, our method learns to predict its relevance to each pixel and derives a See-through-Text Embedding Pixelwise (STEP) heatmap, which reveals segmentation cues of pixel level via the learned visual-textual co-embedding. The ConvRNN performs a top-down approximation by converting the STEP heatmap into a refined one, whereas the improvement is expected from training the network with a classification loss from the ground truth. With the refined heatmap, we update the textual representation of the referring expression by re-evaluating its attention distribution and then compute a new STEP heatmap as the next input to the ConvRNN. Boosting by such collaborative learning, the framework can progressively and simultaneously yield the desired referring segmentation and reasonable attention distribution over the referring sentence. Our method is general and does not rely on, say, the outcomes of object detection from other DNN models, while achieving state-of-the-art performance in all of the four datasets in the experiments.

Recent efforts have shown promising results for person re-identification by designing part-based architectures to allow a neural network to learn discriminative representations from semantically coherent parts. Some efforts use soft attention to reallocate distant outliers to their most similar parts, while others adjust part granularity to incorporate more distant positions for learning the relationships. Others seek to generalize part-based methods by introducing a dropout mechanism on consecutive regions of the feature map to enhance distant region relationships. However, only few prior efforts model the distant or non-local positions of the feature map directly for the person re-ID task. In this paper, we propose a novel attention mechanism to directly model long-range relationships via second-order feature statistics. When combined with a generalized DropBlock module, our method performs equally to or better than state-of-the-art results for mainstream person re-identification datasets, including Market1501, CUHK03, and DukeMTMC-reID.

We present the next generation of MobileNets based on a combination of complementary search techniques as well as a novel architecture design. MobileNetV3 is tuned to mobile phone CPUs through a combination of hardware-aware network architecture search (NAS) complemented by the NetAdapt algorithm and then subsequently improved through novel architecture advances. This paper starts the exploration of how automated search algorithms and network design can work together to harness complementary approaches improving the overall state of the art. Through this process we create two new MobileNet models for release: MobileNetV3-Large and MobileNetV3-Small which are targeted for high and low resource use cases. These models are then adapted and applied to the tasks of object detection and semantic segmentation. For the task of semantic segmentation (or any dense pixel prediction), we propose a new efficient segmentation decoder Lite Reduced Atrous Spatial Pyramid Pooling (LR-ASPP). We achieve new state of the art results for mobile classification, detection and segmentation. MobileNetV3-Large is 3.2% more accurate on ImageNet classification while reducing latency by 20% compared to MobileNetV2. MobileNetV3-Small is 6.6% more accurate compared to a MobileNetV2 model with comparable latency. MobileNetV3-Large detection is over 25% faster at roughly the same accuracy as MobileNetV2 on COCO detection. MobileNetV3-Large LR-ASPP is 34% faster than MobileNetV2 R-ASPP at similar accuracy for Cityscapes segmentation.

While it is trivial for humans to quickly assess the perceptual similarity between two images, the underlying mechanism are thought to be quite complex. Despite this, the most widely adopted perceptual metrics today, such as SSIM and FSIM, are simple, shallow functions, and fail to consider many factors of human perception. Recently, the facial modeling community has observed that the inclusion of both structure and texture has a significant positive benefit for face sketch synthesis (FSS). But how perceptual are these so-called "perceptual features"? Which elements are critical for their success? In this paper, we design a perceptual metric, called Structure Co-Occurrence Texture (Scoot), which simultaneously considers the block-level spatial structure and co-occurrence texture statistics. To test the quality of metrics, we propose three novel meta-measures based on various reliable properties. Extensive experiments verify that our Scoot metric exceeds the performance of prior work. Besides, we built the first largest scale (152k judgments) human-perception-based sketch database that can evaluate how well a metric consistent with human perception. Our results suggest that "spatial structure" and "co-occurrence texture" are two generally applicable perceptual features in face sketch synthesis.

In this paper we propose a neural message passing approach to augment an input 3D indoor scene with new objects matching their surroundings. Given an input, potentially incomplete, 3D scene and a query location, our method predicts a probability distribution over object types that fit well in that location. Our distribution is predicted though passing learned messages in a dense graph whose nodes represent objects in the input scene and edges represent spatial and structural relationships. By weighting messages through an attention mechanism, our method learns to focus on the most relevant surrounding scene context to predict new scene objects. We found that our method significantly outperforms state-of-the-art approaches in terms of correctly predicting objects missing in a scene based on our experiments in the SUNCG dataset. We also demonstrate other applications of our method, including context-based 3D object recognition and iterative scene generation.

Current visual question answering datasets do not consider the rich semantic information conveyed by text within an image. In this work, we present a new dataset, ST-VQA, that aims to highlight the importance of exploiting high-level semantic information present in images as textual cues in the Visual Question Answering process. We use this dataset to define a series of tasks of increasing difficulty for which reading the scene text in the context provided by the visual information is necessary to reason and generate an appropriate answer. We propose a new evaluation metric for these tasks to account both for reasoning errors as well as shortcomings of the text recognition module. In addition we put forward a series of baseline methods, which provide further insight to the newly released dataset, and set the scene for further research.

Visual knowledge bases such as Visual Genome power numerous applications in computer vision, including visual question answering and captioning, but suffer from sparse, incomplete relationships. All scene graph models to date are limited to training on a small set of visual relationships that have thousands of training labels each. Hiring human annotators is expensive, and using textual knowledge base completion methods are incompatible with visual data. In this paper, we introduce a semi-supervised method that assigns probabilistic relationship labels to a large number of unlabeled images using few labeled examples. We analyze visual relationships to suggest two types of image-agnostic features that are used to generate noisy heuristics, whose outputs are aggregated using a factor graph-based generative model. With as few as 10 labeled examples per relationship, the generative model creates enough training data to train any existing state-of-the-art scene graph model. We demonstrate that our method outperforms all baseline approaches on scene graph prediction by5.16 recall@100 for PREDCLS. In our limited label setting, we define a complexity metric for relationships that serves as an indicator (R^2 = 0.778) for conditions under which our method succeeds over transfer learning, the de-facto approach for training with limited labels.

Self-supervised learning aims to learn representations from the data itself without explicit manual supervision. Existing efforts ignore a crucial aspect of self-supervised learning - the ability to scale to large amount of data because self-supervision requires no manual labels. In this work, we revisit this principle and scale two popular self-supervised approaches to 100 million images. We show that by scaling on various axes (including data size and problem 'hardness'), one can largely match or even exceed the performance of supervised pre-training on a variety of tasks such as object detection, surface normal estimation (3D) and visual navigation using reinforcement learning. Scaling these methods also provides many interesting insights into the limitations of current self-supervised techniques and evaluations. We conclude that current self-supervised methods are not 'hard' enough to take full advantage of large scale data and do not seem to learn effective high level semantic representations. We also introduce an extensive benchmark across 9 different datasets and tasks. We believe that such a benchmark along with comparable evaluation settings is necessary to make meaningful progress. Code is at: https://github.com/facebookresearch/fair_self_supervision_benchmark.