CVPR 2021 论文列表

DenseNets introduce concatenation-type skip connections that achieve state-of-the-art accuracy in several computer vision tasks. In this paper, we reveal that the topology of the concatenation-type skip connections is closely related to the gradient propagation which, in turn, enables a predictable behavior of DNNs' test performance. To this end, we introduce a new metric called NN-Mass to quantify how effectively information flows through DNNs. Moreover, we empirically show that NN-Mass also works for other types of skip connections, e.g., for ResNets, Wide-ResNets (WRNs), and MobileNets, which contain addition-type skip connections (i.e., residuals or inverted residuals). As such, for both DenseNet-like CNNs and ResNets/WRNs/MobileNets, our theoretically grounded NN-Mass can identify models with similar accuracy, despite having significantly different size/compute requirements. Detailed experiments on both synthetic and real datasets (e.g., MNIST, CIFAR-10, CIFAR-100, ImageNet) provide extensive evidence for our insights. Finally, the closed-form equation of our NN-Mass enables us to design significantly compressed DenseNets (for CIFAR-10) and MobileNets (for ImageNet) directly at initialization without time-consuming training and/or searching.

This paper proposes a generative adversarial layout refinement network for automated floorplan generation. Our architecture is an integration of a graph-constrained relational GAN and a conditional GAN, where a previously generated layout becomes the next input constraint, enabling iterative refinement. A surprising discovery of our research is that a simple non-iterative training process, dubbed component-wise GT-conditioning, is effective in learning such a generator. The iterative generator further allows us to improve a metric of choice via meta-optimization techniques by controlling when to pass which input constraints during iterative refinement. Our qualitative and quantitative evaluation based on the three standard metrics demonstrate that the proposed system makes significant improvements over the current state-of-the-art, even competitive against the ground-truth floorplans, designed by professional architects. Code, model, and data are available at https://ennauata.github.io/houseganpp/page.html.

Neural Architecture Search (NAS) aims to automatically discover optimal architectures. In this paper, we propose an hourglass-inspired approach (HourNAS) for extremely fast NAS. It is motivated by the fact that the effects of the architecture often proceed from the vital few blocks. Acting like the narrow neck of an hourglass, vital blocks in the guaranteed path from the input to the output of a deep neural network restrict the information flow and influence the network accuracy. The other blocks occupy the major volume of the network and determine the overall network complexity, corresponding to the bulbs of an hourglass. To achieve an extremely fast NAS while preserving the high accuracy, we propose to identify the vital blocks and make them the priority in the architecture search. The search space of those non-vital blocks is further shrunk to only cover the candidates that are affordable under the computational resource constraints. Experimental results on ImageNet show that only using 3 hours (0.1 days) with one GPU, our HourNAS can search an architecture that achieves a 77.0% Top-1 accuracy, which outperforms the state-of-the-art methods.

Existing research on action recognition treats activities as monolithic events occurring in videos. Recently, the benefits of formulating actions as a combination of atomic-actions have shown promise in improving action understanding with the emergence of datasets containing such annotations, allowing us to learn representations capturing this information. However, there remains a lack of studies that extend action composition and leverage multiple viewpoints and multiple modalities of data for representation learning. To promote research in this direction, we introduce Home Action Genome (HOMAGE): a multi-view action dataset with multiple modalities and view-points supplemented with hierarchical activity and atomic action labels together with dense scene composition labels. Leveraging rich multi-modal and multi-view settings, we propose Cooperative Compositional Action Understanding (CCAU), a cooperative learning framework for hierarchical action recognition that is aware of compositional action elements. CCAU shows consistent performance improvements across all modalities. Furthermore, we demonstrate the utility of co-learning compositions in few-shot action recognition by achieving 28.6% mAP with just a single sample.

We present a new pipeline for holistic 3D scene understanding from a single image, which could predict object shape, object pose and scene layout. As it is a highly ill-posed problem, existing methods usually suffer from inaccurate estimation of both shapes and layout especially for the cluttered scene due to the heavy occlusion between objects. We propose to utilize the latest deep implicit representation to solve this challenge. We not only propose an image-based local structured implicit network to improve the object shape estimation, but also refine 3D object pose and scene layout via an novel implicit scene graph neural network that exploits the implicit local object features. A novel physical violation loss is also proposed to avoid incorrect context between objects. Extensive experiments demonstrate that our method outperforms the state-of-the-art methods in terms of object shape, scene layout estimation, and 3D object detection.

The 3D world limits the human body pose and the human body pose conveys information about the surrounding objects. Indeed, from a single image of a person placed in an indoor scene, we as humans are adept at resolving ambiguities of the human pose and room layout through our knowledge of the physical laws and prior perception of the plausible object and human poses. However, few computer vision models fully leverage this fact. In this work, we propose a holistically trainable model that perceives the 3D scene from a single RGB image, estimates the camera pose and the room layout, and reconstructs both human body and object meshes. By imposing a set of comprehensive and sophisticated losses on all aspects of the estimations, we show that our model outperforms existing human body mesh methods and indoor scene reconstruction methods. To the best of our knowledge, this is the first model that outputs both object and human predictions at the mesh level, and performs joint optimization on the scene and human poses.

We present HoHoNet, a versatile and efficient framework for holistic understanding of an indoor 360-degree panorama using a Latent Horizontal Feature (LHFeat). The compact LHFeat flattens the features along the vertical direction and has shown success in modeling per-column modality for room layout reconstruction. HoHoNet advances in two important aspects. First, the deep architecture is redesigned to run faster with improved accuracy. Second, we propose a novel horizon-to-dense module, which relaxes the per-column output shape constraint, allowing per-pixel dense prediction from LHFeat. HoHoNet is fast: It runs at 52 FPS and 110 FPS with ResNet-50 and ResNet-34 backbones respectively, for modeling dense modalities from a high-resolution 512x1024 panorama. HoHoNet is also accurate. On the tasks of layout estimation and semantic segmentation, HoHoNet achieves results on par with current state-of-the-art. On dense depth estimation, HoHoNet outperforms all the prior arts by a large margin.

While generative adversarial networks (GANs) can successfully produce high-quality images, they can be challenging to control. Simplifying GAN-based image generation is critical for their adoption in graphic design and artistic work. This goal has led to significant interest in methods that can intuitively control the appearance of images generated by GANs. In this paper, we present HistoGAN, a color histogram-based method for controlling GAN-generated images' colors. We focus on color histograms as they provide an intuitive way to describe image color while remaining decoupled from domain-specific semantics. Specifically, we introduce an effective modification of the recent StyleGAN architecture [31] to control the colors of GAN-generated images specified by a target color histogram feature. We then describe how to expand HistoGAN to recolor real images. For image recoloring, we jointly train an encoder network along with HistoGAN. The recoloring model, ReHistoGAN, is an unsupervised approach trained to encourage the network to keep the original image's content while changing the colors based on the given target histogram. We show that this histogram-based approach offers a better way to control GAN-generated and real images' colors while producing more compelling results compared to existing alternative strategies.

Sinkhorn divergence has become a very popular metric to compare probability distributions in optimal transport. However, most works resort to Sinkhorn divergence in Euclidean space, which greatly blocks their applications in complex data with nonlinear structure. It is therefore of theoretical demand to empower Sinkhorn divergence with the capability of capturing nonlinear structures. We propose a theoretical and computational framework to bridge this gap. In this paper, we extend Sinkhorn divergence in Euclidean space to the reproducing kernel Hilbert space, which we term "Hilbert Sinkhorn divergence" (HSD).In particular, we can use kernel matrices to derive a closed form expression of HSD that is proved to be a tractable convex optimization problem. We also prove several attractive statistical properties of the proposed HSD, i.e., strong consistency, asymptotic behavior and sample complexity. Empirically, our method yields state-of-the-art performances on image classification and topological data analysis.

While Generative Adversarial Networks (GANs) show increasing performance and the level of realism is becoming indistinguishable from natural images, this also comes with high demands on data and computation. We show that state-of-the-art GAN models -- such as they are being publicly released by researchers and industry -- can be used for a range of applications beyond unconditional image generation. We achieve this by an iterative scheme that also allows gaining control over the image generation process despite the highly non-linear latent spaces of the latest GAN models. We demonstrate that this opens up the possibility to re-use state-of-the-art, difficult to train, pre-trained GANs with a high level of control even if only black-box access is granted. Our work also raises concerns and awareness that the use cases of a published GAN model may well reach beyond the creators' intention, which needs to be taken into account before a full public release. Code is available at https://github.com/a514514772/hijackgan.

Fovea, located in the centre of the retina, is specialized for high-acuity vision. Mimicking the sampling mechanism of the fovea, a retina-inspired camera, named spiking camera, is developed to record the external information with a sampling rate of 40,000 Hz, and outputs asynchronous binary spike streams. Although the temporal resolution of visual information is improved, how to reconstruct the scenes is still a challenging problem. In this paper, we present a novel high-speed image reconstruction model through the short-term plasticity (STP) mechanism of the brain. We derive the relationship between postsynaptic potential regulated by STP and the firing frequency of each pixel. By setting up the STP model at each pixel of the spiking camera, we can infer the scene radiance with the temporal regularity of the spike stream. Moreover, we show that STP can be used to distinguish the static and motion areas and further enhance the reconstruction results. The experimental results show that our methods achieve state-of-the-art performance in both image quality and computing time.

Existing image-to-image translation (I2IT) methods are either constrained to low-resolution images or long inference time due to their heavy computational burden on the convolution of high-resolution feature maps. In this paper, we focus on speeding-up the high-resolution photorealistic I2IT tasks based on closed-form Laplacian pyramid decomposition and reconstruction. Specifically, we reveal that the attribute transformations, such as illumination and color manipulation, relate more to the low-frequency component, while the content details can be adaptively refined on high-frequency components. We consequently propose a Laplacian Pyramid Translation Network (LPTN) to simultaneously perform these two tasks, where we design a lightweight network for translating the low-frequency component with reduced resolution and a progressive masking strategy to efficiently refine the high-frequency ones. Our model avoids most of the heavy computation consumed by processing high-resolution feature maps and faithfully preserves the image details. Extensive experimental results on various tasks demonstrate that the proposed method can translate 4K images in real-time using one normal GPU while achieving comparable transformation performance against existing methods. Datasets and codes are available: https://github.com/csjliang/LPTN.

Single-shot monocular birefractive stereo methods have been used for estimating sparse depth from double refraction over edges. They also obtain an ordinary-ray (o-ray) image concurrently or subsequently through additional post-processing of depth densification and deconvolution. However, when an extraordinary-ray (e-ray) image is restored to acquire stereo images, the existing methods suffer from very severe restoration artifacts in stereo images due to a low signal-to-noise ratio of input e-ray image or depth/deconvolution errors. In this work, we present a novel stereo image restoration network that can restore stereo images directly from a double-refraction image. First, we built a physically faithful birefractive stereo imaging dataset by simulating the double refraction phenomenon with existing RGB-D datasets. Second, we formulated a joint stereo restoration problem that accounts for not only geometric relation between o-/e-ray images but also joint optimization of restoring both stereo images. We trained our model with our birefractive image dataset in an end-to-end manner. Our model restores high-quality stereo images directly from double refraction in real-time, enabling high-quality stereo video using a monocular camera. Our method also allows us to estimate dense depth maps from stereo images using a conventional stereo method. We evaluate the performance of our method experimentally and synthetically with the ground truth. Results validate that our stereo image restoration network outperforms the existing methods with high accuracy. We demonstrate several image-editing applications using our high-quality stereo images and dense depth maps.

Cycle consistency is widely used for face editing. However, we observe that the generator tends to find a tricky way to hide information from the original image to satisfy the constraint of cycle consistency, making it impossible to maintain the rich details (e.g., wrinkles and moles) of nonediting areas. In this work, we propose a simple yet effective method named HifaFace to address the above-mentioned problem from two perspectives. First, we relieve the pressure of the generator to synthesize rich details by directly feeding the high-frequency information of the input image into the end of the generator. Second, we adopt an additional discriminator to encourage the generator to synthesize rich details. Specifically, we apply wavelet transformation to transform the image into multi-frequency domains, among which the high-frequency parts can be used to recover the rich details. We also notice that a fine-grained and wider-range control for the attribute is of great importance for face editing. To achieve this goal, we propose a novel attribute regression loss. Powered by the proposed framework, we achieve high-fidelity and arbitrary face editing, outperforming other state-of-the-art approaches.

Video-based human motion transfer creates video animations of humans following a source motion. Current methods show remarkable results for tightly-clad subjects. However, the lack of temporally consistent handling of plausible clothing dynamics, including fine and high-frequency details, significantly limits the attainable visual quality. We address these limitations for the first time in the literature and present a new framework which performs high-fidelity and temporally-consistent human motion transfer with natural pose-dependent non-rigid deformations, for several types of loose garments. In contrast to the previous techniques, we perform image generation in three subsequent stages: synthesizing human shape, structure, and appearance. Given a monocular RGB video of an actor, we train a stack of recurrent deep neural networks that generate these intermediate representations from 2D poses and their temporal derivatives. Splitting the difficult motion transfer problem into subtasks that are aware of the temporal motion context helps us to synthesize results with plausible dynamics and pose-dependent detail. It also allows artistic control of results by manipulation of individual framework stages. In the experimental results, we significantly outperform the state-of-the-art in terms of video realism. The source code is available at https://graphics.tu-bs.de/publications/kappel2020high-fidelity.

3D video avatars can empower virtual communications by providing compression, privacy, entertainment, and a sense of presence in AR/VR. Best 3D photo-realistic AR/VR avatars driven by video, that can minimize uncanny effects, rely on person-specific models. However, existing person-specific photo-realistic 3D models are not robust to lighting, hence their results typically miss subtle facial behaviors and cause artifacts in the avatar. This is a major drawback for the scalability of these models in communication systems (e.g., Messenger, Skype, FaceTime) and AR/VR. This paper addresses previous limitations by learning a deep learning lighting model, that in combination with a high-quality 3D face tracking algorithm, provides a method for subtle and robust facial motion transfer from a regular video to a 3D photo-realistic avatar. Extensive experimental validation and comparisons to other state-of-the-art methods demonstrate the effectiveness of the proposed framework in real-world scenarios with variability in pose, expression, and illumination. Our project page can be found at https://www.cs.rochester.edu/ cxu22/r/wild-avatar/.

We present a novel two-layer hierarchical reinforcement learning approach equipped with a Goals Relational Graph (GRG) for tackling the partially observable goal-driven task, such as goal-driven visual navigation. Our GRG captures the underlying relations of all goals in the goal space through a Dirichlet-categorical process that facilitates: 1) the high-level network raising a sub-goal towards achieving a designated final goal; 2) the low-level network towards an optimal policy; and 3) the overall system generalizing unseen environments and goals. We evaluate our approach with two settings of partially observable goal-driven tasks -- a grid-world domain and a robotic object search task. Our experimental results show that our approach exhibits superior generalization performance on both unseen environments and new goals.

Learning to model and predict how humans interact with objects while performing an action is challenging, and most of the existing video prediction models are ineffective in modeling complicated human-object interactions. Our work builds on hierarchical video prediction models, which disentangle the video generation process into two stages: predicting a high-level representation, such as pose sequence, and then learning a pose-to-pixels translation model for pixel generation. An action sequence for a human-object interaction task is typically very complicated, involving the evolution of pose, person's appearance, object locations, and object appearances over time. To this end, we propose a Hierarchical Video Prediction model using Relational Layouts. In the first stage, we learn to predict a sequence of layouts. A layout is a high-level representation of the video containing both pose and objects' information for every frame. The layout sequence is learned by modeling the relationships between the pose and objects using relational reasoning and recurrent neural networks. The layout sequence acts as a strong structure prior to the second stage that learns to map the layouts into pixel space. Experimental evaluation of our method on two datasets, UMD-HOI and Bimanual, shows significant improvements in standard video evaluation metrics such as LPIPS, PSNR, and SSIM. We also perform a detailed qualitative analysis of our model to demonstrate various generalizations.

Current approaches to video analysis of human motion focus on raw pixels or keypoints as the basic units of reasoning. We posit that adding higher-level motion primitives, which can capture natural coarser units of motion such as backswing or follow-through, can be used to improve downstream analysis tasks. This higher level of abstraction can also capture key features, such as loops of repeated primitives, that are currently inaccessible at lower levels of representation. We therefore introduce Motion Programs, a neuro-symbolic, program-like representation that expresses motions as a composition of high-level primitives. We also present a system for automatically inducing motion programs from videos of human motion and for leveraging motion programs in video synthesis. Experiments show that motion programs can accurately describe a diverse set of human motions and the inferred programs contain semantically meaningful motion primitives, such as arm swings and jumping jacks. Our representation also benefits downstream tasks such as video interpolation and video prediction and outperforms off-the-shelf models. We further demonstrate how these programs can detect diverse kinds of repetitive motion and facilitate interactive video editing.