CVPR 2022 论文列表

We present Playable Environments - a new representation for interactive video generation and manipulation in space and time. With a single image at inference time, our novel framework allows the user to move objects in 3D while generating a video by providing a sequence of desired actions. The actions are learnt in an unsupervised manner. The camera can be controlled to get the desired viewpoint. Our method builds an environment state for each frame, which can be manipulated by our proposed action module and decoded back to the image space with volumetric rendering. To support diverse appearances of objects, we extend neural radiance fields with style-based modulation. Our method trains on a collection of various monocular videos requiring only the estimated camera parameters and 2D object locations. To set a challenging benchmark, we introduce two large scale video datasets with significant camera movements. As evidenced by our experiments, playable environments enable several creative applications not attainable by prior video synthesis works, including playable 3D video generation, stylization and manipulation. We make our code, demos and datasets publicly available.

We present a novel framework named PlaneMVS for 3D plane reconstruction from multiple input views with known camera poses. Most previous learning-based plane reconstruction methods reconstruct 3D planes from single images, which highly rely on single-view regression and suffer from depth scale ambiguity. In contrast, we reconstruct 3D planes with a multi-view-stereo (MVS) pipeline that takes advantage of multi-view geometry. We decouple plane reconstruction into a semantic plane detection branch and a plane MVS branch. The semantic plane detection branch is based on a single-view plane detection framework but with differences. The plane MVS branch adopts a set of slanted plane hypotheses to replace conventional depth hypotheses to perform plane sweeping strategy and finally learns pixel-level plane parameters and its planar depth map. We present how the two branches are learned in a balanced way, and propose a soft-pooling loss to associate the outputs of the two branches and make them benefit from each other. Extensive experiments on various indoor datasets show that PlaneMVS significantly outperforms state-of-the-art (SOTA) single-view plane reconstruction methods on both plane detection and 3D geometry metrics. Our method even outperforms a set of SOTA learning-based MVS methods thanks to the learned plane priors. To the best of our knowledge, this is the first work on 3D plane reconstruction within an end-to-end MVS framework.

We present PlanarRecon -- a novel framework for globally coherent detection and reconstruction of 3D planes from a posed monocular video. Unlike previous works that detect planes in 2D from a single image, PlanarRecon incrementally detects planes in 3D for each video fragment, which consists of a set of key frames, from a volumetric representation of the scene using neural networks. A learning-based tracking and fusion module is designed to merge planes from previous fragments to form a coherent global plane reconstruction. Such design allows PlanarRecon to integrate observations from multiple views within each fragment and temporal information across different ones, resulting in an accurate and coherent reconstruction of the scene abstraction with low-polygonal geometry. Experiments show that the proposed approach achieves state-of-the-art performances on the ScanNet dataset while being real-time.

Blind deblurring has attracted much interest with its wide applications in reality. The blind deblurring problem is usually solved by estimating the intermediate kernel and the intermediate image alternatively, which will finally converge to the blurring kernel of the observed image. Numerous works have been proposed to obtain intermediate images with fewer undesirable artifacts by designing delicate regularization on the latent solution. However, these methods still fail while dealing with images containing saturations and large blurs. To address this problem, we propose an intermediate image correction method which utilizes Bayes posterior estimation to screen through the intermediate image and exclude those unfavorable pixels to reduce their influence for kernel estimation. Extensive experiments have proved that the proposed method can effectively improve the accuracy of the final derived kernel against the state-of-the-art methods on benchmark datasets by both quantitative and qualitative comparisons.

In real-world applications of machine learning, reliable and safe systems must consider measures of performance beyond standard test set accuracy. These other goals include out-of-distribution (OOD) robustness, prediction consistency, resilience to adversaries, calibrated uncertainty estimates, and the ability to detect anomalous inputs. However, improving performance towards these goals is often a balancing act that today's methods cannot achieve without sacrificing performance on other safety axes. For instance, adversarial training improves adversarial robustness but sharply degrades other classifier performance metrics. Similarly, strong data augmentation and regularization techniques often improve OOD robustness but harm anomaly detection, raising the question of whether a Pareto improvement on all existing safety measures is possible. To meet this challenge, we design a new data augmentation strategy utilizing the natural structural complexity of pictures such as fractals, which outperforms numerous baselines, is near Pareto-optimal, and roundly improves safety measures.

We propose a pipeline to generate Neural Radiance Fields (NeRF) of an object or a scene of a specific class, conditioned on a single input image. This is a challenging task, as training NeRF requires multiple views of the same scene, coupled with corresponding poses, which are hard to obtain. Our method is based on pi-GAN, a generative model for unconditional 3D-aware image synthesis, which maps random latent codes to radiance fields of a class of objects. We jointly optimize (1) the pi-GAN objective to utilize its high-fidelity 3D-aware generation and (2) a carefully designed reconstruction objective. The latter includes an encoder coupled with pi-GAN generator to form an auto-encoder. Unlike previous few-shot NeRF approaches, our pipeline is unsupervised, capable of being trained with independent images without 3D, multi-view, or pose supervision. Applications of our pipeline include 3d avatar generation, object-centric novel view synthesis with a single input image, and 3d-aware super-resolution, to name a few.

The rise of deep neural networks has led to several breakthroughs for semantic segmentation. In spite of this, a model trained on source domain often fails to work properly in new challenging domains, that is directly concerned with the generalization capability of the model. In this paper, we present a novel memory-guided domain generalization method for semantic segmentation based on meta-learning framework. Especially, our method abstracts the conceptual knowledge of semantic classes into categorical memory which is constant beyond the domains. Upon the meta-learning concept, we repeatedly train memory-guided networks and simulate virtual test to 1) learn how to memorize a domain-agnostic and distinct information of classes and 2) offer an externally settled memory as a class-guidance to reduce the ambiguity of representation in the test data of arbitrary unseen domain. To this end, we also propose memory divergence and feature cohesion losses, which encourage to learn memory reading and update processes for category-aware domain generalization. Extensive experiments for semantic segmentation demonstrate the superior generalization capability of our method over state-of-the-art works on various benchmarks.

This paper presents a novel Physically-guided Disentangled Implicit Rendering (PhyDIR) framework for high-fidelity 3D face modeling. The motivation comes from two observations: widely-used graphics renderers yield excessive approximations against photo-realistic imaging, while neural rendering methods are highly entangled to perceive 3D-aware operations. Hence, we learn to disentangle the implicit rendering via explicit physical guidance, meanwhile guarantee the properties of (1) 3D-aware comprehension and (2) high-reality imaging. For the former one, PhyDIR explicitly adopts 3D shading and rasterizing modules to control the renderer, which disentangles the lighting, facial shape and view point from neural reasoning. Specifically, PhyDIR proposes a novel multi-image shading strategy to compensate the monocular limitation, so that the lighting variations are accessible to the neural renderer. For the latter one, PhyDIR learns the face-collection implicit texture to avoid ill-posed intrinsic factorization, then leverages a series of consistency losses to constrain the robustness. With the disentangled method, we make 3D face modeling benefit from both kinds of rendering strategies. Extensive experiments on benchmarks show that PhyDIR obtains superior performance than state-of-the-art explicit/implicit methods, on both geometry/texture modeling.

Learning-based image dehazing methods have achieved marvelous progress during the past few years. On one hand, most approaches heavily rely on synthetic data and may face difficulties to generalize well in real scenes, due to the huge domain gap between synthetic and real images. On the other hand, very few works have considered the varicolored haze, caused by chromatic casts in real scenes. In this work, our goal is to handle the new task: real-world varicolored haze removal. To this end, we propose a physically disentangled joint intra- and inter-domain adaptation paradigm, in which intra-domain adaptation focuses on color correction and inter-domain procedure transfers knowledge between synthetic and real domains. We first learn to physically disentangle haze images into three components complying with the scattering model: background, transmission map, and atmospheric light. Since haze color is determined by atmospheric light, we perform intra-domain adaptation by specifically translating atmospheric light from varicolored space to unified color-balanced space, and then reconstructing color-balanced haze image through the scattering model. Consequently, we perform inter-domain adaptation between the synthetic and real images by mutually exchanging the background and other two components. Then we can reconstruct both identity and domain-translated haze images with self-consistency and adversarial loss. Extensive experiments demonstrate the superiority of the proposed method over the state-of-the-art for real varicolored image dehazing.

We introduce a novel approach for generative 3D modeling that explicitly encourages the physical and thus functional consistency of the generated shapes. To this end, we advocate the use of online physical simulation as part of learning a generative model. Unlike previous related methods, our approach is trained end-to-end with a fully differentiable physical simulator in the training loop. We accomplish this by leveraging recent advances in differentiable programming, and introducing a fully differentiable point-based physical simulation layer, which accurately evaluates the shape's stability when subjected to gravity. We then incorporate this layer in a signed distance function (SDF) shape decoder. By augmenting a conventional SDF decoder with our simulation layer, we demonstrate through extensive experiments that online physical simulation improves the accuracy, visual plausibility and physical validity of the resulting shapes, while requiring no additional data or annotation effort.

Motion capture from sparse inertial sensors has shown great potential compared to image-based approaches since occlusions do not lead to a reduced tracking quality and the recording space is not restricted to be within the viewing frustum of the camera. However, capturing the motion and global position only from a sparse set of inertial sensors is inherently ambiguous and challenging. In consequence, recent state-of-the-art methods can barely handle very long period motions, and unrealistic artifacts are common due to the unawareness of physical constraints. To this end, we present the first method which combines a neural kinematics estimator and a physics-aware motion optimizer to track body motions with only 6 inertial sensors. The kinematics module first regresses the motion status as a reference, and then the physics module refines the motion to satisfy the physical constraints. Experiments demonstrate a clear improvement over the state of the art in terms of capture accuracy, temporal stability, and physical correctness.

Remote photoplethysmography (rPPG), which aims at measuring heart activities and physiological signals from facial video without any contact, has great potential in many applications. Recent deep learning approaches focus on mining subtle rPPG clues using convolutional neural networks with limited spatio-temporal receptive fields, which neglect the long-range spatio-temporal perception and interaction for rPPG modeling. In this paper, we propose the PhysFormer, an end-to-end video transformer based architecture, to adaptively aggregate both local and global spatio-temporal features for rPPG representation enhancement. As key modules in PhysFormer, the temporal difference transformers first enhance the quasi-periodic rPPG features with temporal difference guided global attention, and then refine the local spatio-temporal representation against interference. Furthermore, we also propose the label distribution learning and a curriculum learning inspired dynamic constraint in frequency domain, which provide elaborate supervisions for PhysFormer and alleviate overfitting. Comprehensive experiments are performed on four benchmark datasets to show our superior performance on both intra- and cross-dataset testings. One highlight is that, unlike most transformer networks needed pretraining from large-scale datasets, the proposed PhysFormer can be easily trained from scratch on rPPG datasets, which makes it promising as a novel transformer baseline for the rPPG community. The codes will be released soon.

Inverse rendering of complex material such as glossy, metal and mirror material is a long-standing ill-posed problem in this area, which has not been well solved. Previous approaches cannot tackle them well due to simplified BRDF and unsuitable illumination representations. In this paper, we present PhyIR, a neural inverse rendering method with a more completed SVBRDF representation and a physics-based in-network rendering layer, which can handle complex material and incorporate physical constraints by re-rendering realistic and detailed specular reflectance. Our framework estimates geometry, material and Spatially-Coherent (SC) illumination from a single indoor panorama. Due to the lack of panoramic datasets with completed SVBRDF and full-spherical light probes, we introduce an artist-designed dataset named FutureHouse with high-quality geometry, SVBRDF and per-pixel Spatially-Varying (SV) lighting. To ensure the coherence of SV lighting, a novel SC loss is proposed. Extensive experiments on both synthetic and real-world data show that the proposed method outperforms the state-of-the-arts quantitatively and qualitatively, and is able to produce photorealistic results for a number of applications such as dynamic virtual object insertion.

We present PHORHUM, a novel, end-to-end trainable, deep neural network methodology for photorealistic 3D human reconstruction given just a monocular RGB image. Our pixel-aligned method estimates detailed 3D geometry and, for the first time, the unshaded surface color together with the scene illumination. Observing that 3D supervision alone is not sufficient for high fidelity color reconstruction, we introduce patch-based rendering losses that enable reliable color reconstruction on visible parts of the human, and detailed and plausible color estimation for the non-visible parts. Moreover, our method specifically addresses methodological and practical limitations of prior work in terms of representing geometry, albedo, and illumination effects, in an end-to-end model where factors can be effectively disentangled. In extensive experiments, we demonstrate the versatility and robustness of our approach. Our state-of-the-art results validate the method qualitatively and for different metrics, for both geometric and color reconstruction.

Most indoor 3D scene reconstruction methods focus on recovering 3D geometry and scene layout. In this work, we go beyond this to propose PhotoScene, a framework that takes input image(s) of a scene along with approximately aligned CAD geometry (either reconstructed automatically or manually specified) and builds a photorealistic digital twin with high-quality materials and similar lighting. We model scene materials using procedural material graphs; such graphs represent photorealistic and resolution-independent materials. We optimize the parameters of these graphs and their texture scale and rotation, as well as the scene lighting to best match the input image via a differentiable rendering layer. We evaluate our technique on objects and layout reconstructions from ScanNet, SUN RGB-D and stock photographs, and demonstrate that our method reconstructs high-quality, fully relightable 3D scenes that can be re-rendered under arbitrary viewpoints, zooms and lighting.

Object pose estimation is crucial for robotic applications and augmented reality. Beyond instance level 6D object pose estimation methods, estimating category-level pose and shape has become a promising trend. As such, a new research field needs to be supported by well-designed datasets. To provide a benchmark with high-quality ground truth annotations to the community, we introduce a multimodal dataset for category-level object pose estimation with photometrically challenging objects termed PhoCaL. PhoCaL comprises 60 high quality 3D models of household objects over 8 categories including highly reflective, transparent and symmetric objects. We developed a novel robot-supported multi-modal (RGB, depth, polarisation) data acquisition and annotation process. It ensures sub-millimeter accuracy of the pose for opaque textured, shiny and transparent objects, no motion blur and perfect camera synchronisation. To set a benchmark for our dataset, state-of-the-art RGB-D and monocular RGB methods are evaluated on the challenging scenes of PhoCaL.

Consistency learning using input image, feature, or network perturbations has shown remarkable results in semi-supervised semantic segmentation, but this approach can be seriously affected by inaccurate predictions of unlabelled training images. There are two consequences of these inaccurate predictions: 1) the training based on the "strict" cross-entropy (CE) loss can easily overfit prediction mistakes, leading to confirmation bias; and 2) the perturbations applied to these inaccurate predictions will use potentially erroneous predictions as training signals, degrading consistency learning. In this paper, we address the prediction accuracy problem of consistency learning methods with novel extensions of the mean-teacher (MT) model, which include a new auxiliary teacher, and the replacement of MT's mean square error (MSE) by a stricter confidence-weighted cross-entropy (Conf-CE) loss. The accurate prediction by this model allows us to use a challenging combination of network, input data and feature perturbations to improve the consistency learning generalisation, where the feature perturbations consist of a new adversarial perturbation. Results on public benchmarks show that our approach achieves remarkable improvements over the previous SOTA methods in the field.

Personalized image aesthetics assessment (PIAA) is challenging due to its highly subjective nature. People's aesthetic tastes depend on diversified factors, including image characteristics and subject characters. The existing PIAA databases are limited in terms of annotation diversity, especially the subject aspect, which can no longer meet the increasing demands of PIAA research. To solve the dilemma, we conduct so far, the most comprehensive subjective study of personalized image aesthetics and introduce a new Personalized image Aesthetics database with Rich Attributes (PARA), which consists of 31,220 images with annotations by 438 subjects. PARA features wealthy annotations, including 9 image-oriented objective attributes and 4 human-oriented subjective attributes. In addition, desensitized subject information, such as personality traits, is also provided to support study of PIAA and user portraits. A comprehensive analysis of the annotation data is provided and statistic study indicates that the aesthetic preferences can be mirrored by proposed subjective attributes. We also propose a conditional PIAA model by utilizing subject information as conditional prior. Experimental results indicate that the conditional PIAA model can outperform the control group, which is also the first attempt to demonstrate how image aesthetics and subject characters interact to produce the intricate personalized tastes on image aesthetics. We believe the database and the associated analysis would be useful for conducting next-generation PIAA study. The project page of PARA can be found at https://cv-datasets.institutecv.com/#/data-sets.

Knowledge distillation has shown great effectiveness for improving neural architecture search (NAS). Mutual knowledge distillation (MKD), where a group of models mutually generate knowledge to train each other, has achieved promising results in many applications. In existing MKD methods, mutual knowledge distillation is performed between models without scrutiny: a worse-performing model is allowed to generate knowledge to train a better-performing model, which may lead to collective failures. To address this problem, we propose a performance-aware MKD (PAMKD) approach for NAS, where knowledge generated by model A is allowed to train model B only if the performance of A is better than B. We propose a three-level optimization framework to formulate PAMKD, where three learning stages are performed end-to-end: 1) each model trains an initial model independently; 2) the initial models are evaluated on a validation set and better-performing models generate knowledge to train worse-performing models; 3) architectures are updated by minimizing a validation loss. Experimental results on a variety of datasets demonstrate that our method is effective.