CVPR 2019 论文列表

Binarized Neural Networks (BNNs) can significantly reduce the inference latency and energy consumption in resource-constrained devices due to their pure-logical computation and fewer memory accesses. However, training BNNs is difficult since the activation flow encounters degeneration, saturation, and gradient mismatch problems. Prior work alleviates these issues by increasing activation bits and adding floating-point scaling factors, thereby sacrificing BNN's energy efficiency. In this paper, we propose to use distribution loss to explicitly regularize the activation flow, and develop a framework to systematically formulate the loss. Our experiments show that the distribution loss can consistently improve the accuracy of BNNs without losing their energy benefits. Moreover, equipped with the proposed regularization, BNN training is shown to be robust to the selection of hyper-parameters including optimizer and learning rate.

We consider the face recognition task where facial images of the same identity (person) is expected to be closer in the representation space, while different identities be far apart. Several recent studies encourage the intra-class compactness by developing loss functions that penalize the variance of representations of the same identity. In this paper, we propose the `exclusive regularization' that focuses on the other aspect of discriminability -- the inter-class separability, which is neglected in many recent approaches. The proposed method, named RegularFace, explicitly distances identities by penalizing the angle between an identity and its nearest neighbor, resulting in discriminative face representations. Our method has intuitive geometric interpretation and presents unique benefits that are absent in previous works. Quantitative comparisons against prior methods on several open benchmarks demonstrate the superiority of our method. In addition, our method is easy to implement and requires only a few lines of python code on modern deep learning frameworks.

Region anchors are the cornerstone of modern object detection techniques. State-of-the-art detectors mostly rely on a dense anchoring scheme, where anchors are sampled uniformly over the spatial domain with a predefined set of scales and aspect ratios. In this paper, we revisit this foundational stage. Our study shows that it can be done much more effectively and efficiently. Specifically, we present an alternative scheme, named Guided Anchoring, which leverages semantic features to guide the anchoring. The proposed method jointly predicts the locations where the center of objects of interest are likely to exist as well as the scales and aspect ratios at different locations. On top of predicted anchor shapes, we mitigate the feature inconsistency with a feature adaption module. We also study the use of high-quality proposals to improve detection performance. The anchoring scheme can be seamlessly integrated into proposal methods and detectors. With Guided Anchoring, we achieve 9.1% higher recall on MS COCO with 90% fewer anchors than the RPN baseline. We also adopt Guided Anchoring in Fast R-CNN, Faster R-CNN and RetinaNet, respectively improving the detection mAP by 2.2%, 2.7% and 1.2%. Code is available at https://github.com/open-mmlab/mmdetection.

In this paper, we address the separation of reflective and fluorescent components in RGB images taken under narrow-band light sources such as LEDs. First, we show that the fluorescent color per pixel can be estimated from at least two images under different light source colors, because the observed color at a surface point is represented by a convex combination of the light source color and the illumination-invariant fluorescent color. Second, we propose a method for robustly estimating the fluorescent color via MAP estimation by taking the prior knowledge with respect to fluorescent colors into consideration. We conducted a number of experiments by using both synthetic and real images, and confirmed that our proposed method works better than the closely related state-of-the-art method and enables us to separate reflective and fluorescent components even from a single image. Furthermore, we demonstrate that our method is effective for applications such as image-based material editing and relighting.

Reflection removal is the challenging problem of removing unwanted reflections that occur when imaging a scene that is behind a pane of glass. In this paper, we show that most cameras have an overlooked mechanism that can greatly simplify this task. Specifically, modern DLSR and smartphone cameras use dual pixel (DP) sensors that have two photodiodes per pixel to provide two sub-aperture views of the scene from a single captured image. "Defocus-disparity" cues, which are natural by-products of the DP sensor encoded within these two sub-aperture views, can be used to distinguish between image gradients belonging to the in-focus background and those caused by reflection interference. This gradient information can then be incorporated into an optimization framework to recover the background layer with higher accuracy than currently possible from the single captured image. As part of this work, we provide the first image dataset for reflection removal consisting of the sub-aperture views from the DP sensor.

Nowadays, the majority of state of the art monocular depth estimation techniques are based on supervised deep learning models. However, collecting RGB images with associated depth maps is a very time consuming procedure. Therefore, recent works have proposed deep architectures for addressing the monocular depth prediction task as a reconstruction problem, thus avoiding the need of collecting ground-truth depth. Following these works, we propose a novel self-supervised deep model for estimating depth maps. Our framework exploits two main strategies: refinement via cycle-inconsistency and distillation. Specifically, first a student network is trained to predict a disparity map such as to recover from a frame in a camera view the associated image in the opposite view. Then, a backward cycle network is applied to the generated image to re-synthesize back the input image, estimating the opposite disparity. A third network exploits the inconsistency between the original and the reconstructed input frame in order to output a refined depth map. Finally, knowledge distillation is exploited, such as to transfer information from the refinement network to the student. Our extensive experimental evaluation demonstrate the effectiveness of the proposed framework which outperforms state of the art unsupervised methods on the KITTI benchmark.

The goal of MRI reconstruction is to restore a high fidelity image from partially observed measurements. This partial view naturally induces reconstruction uncertainty that can only be reduced by acquiring additional measurements. In this paper, we present a novel method for MRI reconstruction that, at inference time, dynamically selects the measurements to take and iteratively refines the prediction in order to best reduce the reconstruction error and, thus, its uncertainty. We validate our method on a large scale knee MRI dataset, as well as on ImageNet. Results show that (1) our system successfully outperforms active acquisition baselines; (2) our uncertainty estimates correlate with error maps; and (3) our ResNet-based architecture surpasses standard pixel-to-pixel models in the task of MRI reconstruction. The proposed method not only shows high-quality reconstructions but also paves the road towards more applicable solutions for accelerating MRI.

Visual dialog is a challenging vision-language task, which requires the agent to answer multi-round questions about an image. It typically needs to address two major problems: (1) How to answer visually-grounded questions, which is the core challenge in visual question answering (VQA); (2) How to infer the co-reference between questions and the dialog history. An example of visual co-reference is: pronouns (e.g., "they") in the question (e.g., "Are they on or off?") are linked with nouns (e.g., "lamps") appearing in the dialog history (e.g., "How many lamps are there?") and the object grounded in the image. In this work, to resolve the visual co-reference for visual dialog, we propose a novel attention mechanism called Recursive Visual Attention (RvA). Specifically, our dialog agent browses the dialog history until the agent has sufficient confidence in the visual co-reference resolution, and refines the visual attention recursively. The quantitative and qualitative experimental results on the large-scale VisDial v0.9 and v1.0 datasets demonstrate that the proposed RvA not only outperforms the state-of-the-art methods, but also achieves reasonable recursion and interpretable attention maps without additional annotations. The code is available at https://github.com/yuleiniu/rva.

Recurrent neural networks (RNNs) are widely used for sequential data processing. Recent state-of-the-art video deblurring methods bank on convolutional recurrent neural network architectures to exploit the temporal relationship between neighboring frames. In this work, we aim to improve the accuracy of recurrent models by adapting the hidden states transferred from past frames to the frame being processed so that the relations between video frames could be better used. We iteratively update the hidden state via re-using RNN cell parameters before predicting an output deblurred frame. Since we use existing parameters to update the hidden state, our method improves accuracy without additional modules. As the architecture remains the same regardless of iteration number, fewer iteration models can be considered as a partial computational path of the models with more iterations. To take advantage of this property, we employ a stochastic method to optimize our iterative models better. At training time, we randomly choose the iteration number on the fly and apply a regularization loss that favors less computation unless there are considerable reconstruction gains. We show that our method exhibits state-of-the-art video deblurring performance while operating in real-time speed.

Deep learning-based, single-view depth estimation methods have recently shown highly promising results. However, such methods ignore one of the most important features for determining depth in the human vision system, which is motion. We propose a learning-based, multi-view dense depth map and odometry estimation method that uses Recurrent Neural Networks (RNN) and trains utilizing multi-view image reprojection and forward-backward flow-consistency losses. Our model can be trained in a supervised or even unsupervised mode. It is designed for depth and visual odometry estimation from video where the input frames are temporally correlated. However, it also generalizes to single-view depth estimation. Our method produces superior results to the state-of-the-art approaches for single-view and multi-view learning-based depth estimation on the KITTI driving dataset.

Deep learning has recently demonstrated its excellent performance for multi-view stereo (MVS). However, one major limitation of current learned MVS approaches is the scalability: the memory-consuming cost volume regularization makes the learned MVS hard to be applied to high-resolution scenes. In this paper, we introduce a scalable multi-view stereo framework based on the recurrent neural network. Instead of regularizing the entire 3D cost volume in one go, the proposed Recurrent Multi-view Stereo Network (R-MVSNet) sequentially regularizes the 2D cost maps along the depth direction via the gated recurrent unit (GRU). This reduces dramatically the memory consumption and makes high-resolution reconstruction feasible. We first show the state-of-the-art performance achieved by the proposed R-MVSNet on the recent MVS benchmarks. Then, we further demonstrate the scalability of the proposed method on several large-scale scenarios, where previous learned approaches often fail due to the memory constraint. Code is available at https://github.com/YoYo000/MVSNet.

We proposed a novel architecture for the problem of video super-resolution. We integrate spatial and temporal contexts from continuous video frames using a recurrent encoder-decoder module, that fuses multi-frame information with the more traditional, single frame super-resolution path for the target frame. In contrast to most prior work where frames are pooled together by stacking or warping, our model, the Recurrent Back-Projection Network (RBPN) treats each context frame as a separate source of information. These sources are combined in an iterative refinement framework inspired by the idea of back-projection in multiple-image super-resolution. This is aided by explicitly representing estimated inter-frame motion with respect to the target, rather than explicitly aligning frames. We propose a new video super-resolution benchmark, allowing evaluation at a larger scale and considering videos in different motion regimes. Experimental results demonstrate that our RBPN is superior to existing methods on several datasets.

Crowd counting is a new frontier in computer vision with far-reaching applications particularly in social safety management. A majority of existing works adopt a methodology that first estimates a person-density map and then calculates integral over this map to obtain the final count. As noticed by several prior investigations, the learned density map can significantly deviate from the true person density even though the final reported count is precise. This implies that the density map is unreliable for localizing crowd. To address this issue, this work proposes a novel framework that simultaneously solving two inherently related tasks - crowd counting and localization. The contributions are several-fold. First, our formulation is based on a crucial observation that localization tends to be inaccurate at high-density regions, and increasing the resolution is an effective albeit simple solution for improving localization. We thus propose Recurrent Attentive Zooming Network, which recurrently detects ambiguous image region and zooms it into high resolution for re-inspection. Second, the two tasks of counting and localization mutually reinforce each other. We propose an adaptive fusion scheme that effectively elevates the performance. Finally, a well-defined evaluation metric is proposed for the rarely-explored localization task. We conduct comprehensive evaluations on several crowd benchmarks, including the newly-developed large-scale UCF-QNRF dataset and demonstrate superior advantages over state-of-the-art methods.

In this paper, we address the large-scale object detection problem with thousands of categories, which poses severe challenges due to long-tail data distributions, heavy occlusions, and class ambiguities. However, the dominant object detection paradigm is limited by treating each object region separately without considering crucial semantic dependencies among objects. In this work, we introduce a novel Reasoning-RCNN to endow any detection networks the capability of adaptive global reasoning over all object regions by exploiting diverse human commonsense knowledge. Instead of only propagating the visual features on the image directly, we evolve the high-level semantic representations of all categories globally to avoid distracted or poor visual features in the image. Specifically, built on feature representations of basic detection network, the proposed network first generates a global semantic pool by collecting the weights of previous classification layer for each category, and then adaptively enhances each object features via attending different semantic contexts in the global semantic pool. Rather than propagating information from all semantic information that may be noisy, our adaptive global reasoning automatically discovers most relative categories for feature evolving. Our Reasoning-RCNN is light-weight and flexible enough to enhance any detection backbone networks, and extensible for integrating any knowledge resources. Solid experiments on object detection benchmarks show the superiority of our Reasoning-RCNN, e.g. achieving around 16% improvement on VisualGenome, 37% on ADE in terms of mAP and 15% improvement on COCO.

We propose a novel model to address the task of Visual Dialog which exhibits complex dialog structures. To obtain a reasonable answer based on the current question and the dialog history, the underlying semantic dependencies between dialog entities are essential. In this paper, we explicitly formalize this task as inference in a graphical model with partially observed nodes and unknown graph structures (relations in dialog). The given dialog entities are viewed as the observed nodes. The answer to a given question is represented by a node with missing value. We first introduce an Expectation Maximization algorithm to infer both the underlying dialog structures and the missing node values (desired answers). Based on this, we proceed to propose a differentiable graph neural network (GNN) solution that approximates this process. Experiment results on the VisDial and VisDial-Q datasets show that our model outperforms comparative methods. It is also observed that our method can infer the underlying dialog structure for better dialog reasoning.

Deep convolutional neural networks trained end-to-end are the state-of-the-art methods to regress dense disparity maps from stereo pairs. These models, however, suffer from a notable decrease in accuracy when exposed to scenarios significantly different from the training set (e.g., real vs synthetic images, etc.). We argue that it is extremely unlikely to gather enough samples to achieve effective training/tuning in any target domain, thus making this setup impractical for many applications. Instead, we propose to perform unsupervised and continuous online adaptation of a deep stereo network, which allows for preserving its accuracy in any environment. However, this strategy is extremely computationally demanding and thus prevents real-time inference. We address this issue introducing a new lightweight, yet effective, deep stereo architecture, Modularly ADaptive Network(MADNet), and developing a Modular ADaptation (MAD) algorithm, which independently trains sub-portions of the network. By deploying MADNet together with MAD we introduce the first real-time self-adaptive deep stereo system enabling competitive performance on heterogeneous datasets. Our code is publicly available at https://github.com/CVLAB-Unibo/Real-time-self-adaptive-deep-stereo.

A well-trained Convolutional Neural Network can easily be pruned without significant loss of performance. This is because of unnecessary overlap in the features captured by the network's filters. Innovations in network architecture such as skip/dense connections and inception units have mitigated this problem to some extent, but these improvements come with increased computation and memory requirements at run-time. We attempt to address this problem from another angle - not by changing the network structure but by altering the training method. We show that by temporarily pruning and then restoring a subset of the model's filters, and repeating this process cyclically, overlap in the learned features is reduced, producing improved generalization. We show that the existing model-pruning criteria are not optimal for selecting filters to prune in this context, and introduce inter-filter orthogonality as the ranking criteria to determine under-expressive filters. Our method is applicable both to vanilla convolutional networks and more complex modern architectures, and improves the performance across a variety of tasks, especially when applied to smaller networks.

This work studies the unsupervised re-ranking procedure for object retrieval and person re-identification with a specific concentration on an ensemble of multiple metrics (or similarities). While the re-ranking step is involved by running a diffusion process on the underlying data manifolds, the fusion step can leverage the complementarity of multiple metrics. We give a comprehensive summary of existing fusion with diffusion strategies, and systematically analyze their pros and cons. Based on the analysis, we propose a unified yet robust algorithm which inherits their advantages and discards their disadvantages. Hence, we call it Unified Ensemble Diffusion (UED). More interestingly, we derive that the inherited properties indeed stem from a theoretical framework, where the relevant works can be elegantly summarized as special cases of UED by imposing additional constraints on the objective function and varying the solver of similarity propagation. Extensive experiments with 3D shape retrieval, image retrieval and person re-identification demonstrate that the proposed framework outperforms the state of the arts, and at the same time suggest that re-ranking via metric fusion is a promising tool to further improve the retrieval performance of existing algorithms.

We propose a new deep architecture for person re-identification (re-id). While re-id has seen much recent progress, spatial localization and view-invariant representation learning for robust cross-view matching remain key, unsolved problems. We address these questions by means of a new attention-driven Siamese learning architecture, called the Consistent Attentive Siamese Network. Our key innovations compared to existing, competing methods include (a) a flexible framework design that produces attention with only identity labels as supervision, (b) explicit mechanisms to enforce attention consistency among images of the same person, and (c) a new Siamese framework that integrates attention and attention consistency, producing principled supervisory signals as well as the first mechanism that can explain the reasoning behind the Siamese framework's predictions. We conduct extensive evaluations on the CUHK03-NP, DukeMTMC-ReID, and Market-1501 datasets and report competitive performance.