We present locality-sensitive avatar, a neural radiance field (NeRF) based network to learn human motions from monocular videos. To this end, we estimate a canonical representation between different frames of a video with a non-linear mapping from observation to canonical space, which we decompose into a skeletal rigid motion and a non-rigid counterpart. Our key contribution is to retain fine-grained details by modeling the non-rigid part with a graph neural network (GNN) that keeps the pose information local to neighboring body parts. Compared to former canonical representation based methods which solely operate on the coordinate space of a whole shape, our locality-sensitive motion modeling can reproduce both realistic shape contours and vivid fine-grained details. We evaluate on ZJU-MoCap, SynWild, ActorsHQ, MVHumanNet and various outdoor videos. The experiments reveal that with the locality sensitive deformation to canonical feature space, we are the first to achieve state-of-the-art results across novel view synthesis, novel pose animation and 3D shape reconstruction simultaneously. Our code is available at https://github.com/ChunjinSong/lsavatar.

Abstract For reconstructing high‐fidelity human 3D models from monocular videos, it is crucial to maintain consistent large‐scale body shapes along with finely matched subtle wrinkles. This paper explores how per‐frame rendering results can be factorized into a pose‐independent component and a corresponding pose‐dependent counterpart to facilitate frame consistency at multiple scales. Pose adaptive texture features are further improved by restricting the frequency bands of these two components. Pose‐independent outputs are expected to be low‐frequency, while high‐frequency information is linked to pose‐dependent factors. We implement this with a dual‐branch network. The first branch takes coordinates in the canonical space as input, while the second one additionally considers features outputted by the first branch and pose information of each frame. A final network integrates the information predicted by both branches and utilizes volume rendering to generate photo‐realistic 3D human images. Through experiments, we demonstrate that our method consistently surpasses all state‐of‐the‐art methods in preserving high‐frequency details and ensuring consistent body contours. Our code is accessible at https://github.com/ChunjinSong/facavatar .

Human motion capture either requires multi-camera systems or is unreliable when using single-view input due to depth ambiguities. Meanwhile, mirrors are readily available in urban environments and form an affordable alternative by recording two views with only a single camera. However, the mirror setting poses the additional challenge of handling occlusions of real and mirror image. Going beyond existing mirror approaches for 3D human pose estimation, we utilize mirrors for learning a complete body model, including shape and dense appearance. Our main contributions are extending articulated neural radiance fields to include a notion of a mirror, making it sample-efficient over potential occlusion regions. Together, our contributions realize a consumer-level 3D motion capture system that starts from off-the-shelf 2D poses by automatically calibrating the camera, estimating mirror orientation, and subsequently lifting 2D keypoint detections to 3D skeleton pose that is used to condition the mirror-aware NeRF. We empirically demonstrate the benefit of learning a body model and accounting for occlusion in challenging mirror scenes. The project is available at: https://danielajisafe.github.io/mirror-aware-neural-humans/.

It is now possible to estimate 3D human pose from monocular images with off-the-shelf 3D pose estimators. However, many practical applications require fine-grained absolute pose information for which multi-view cues and camera calibration are necessary. Such multi-view recordings are laborious because they require manual calibration, and are expensive when using dedicated hardware. Our goal is full automation, which includes temporal synchronization, as well as intrinsic and extrinsic camera calibration. This is done by using persons in the scene as the calibration objects. Existing methods either address only synchronization or calibration, assume one of the former as input, or have significant limitations. A common limitation is that they only consider single persons, which eases correspondence finding. We attain this generality by partitioning the high-dimensional time and calibration space into a cascade of subspaces and introduce tailored algorithms to optimize each efficiently and robustly. The outcome is an easy-to-use, flexible, and robust motion capture toolbox that we release to enable scientific applications, which we demonstrate on diverse multi-view benchmarks. Project website: https://github.com/tangytoby/CasCalib.

It is now possible to reconstruct dynamic human motion and shape from a sparse set of cameras using Neural Radiance Fields (NeRF) driven by an underlying skeleton. However, a challenge remains to model the deformation of cloth and skin in relation to skeleton pose. Unlike existing avatar models that are learned implicitly or rely on a proxy surface, our approach is motivated by the observation that different poses necessitate unique frequency assignments. Neglecting this distinction yields noisy artifacts in smooth areas or blurs fine-grained texture and shape details in sharp regions. We develop a two-branch neural network that is adaptive and explicit in the frequency domain. The first branch is a graph neural network that models correlations among body parts locally, taking skeleton pose as input. The second branch combines these correlation features to a set of global frequencies and then modulates the feature encoding. Our experiments demonstrate that our network outperforms state-of-the-art methods in terms of preserving details and generalization capabilities. Our code is available at https://github.com/ChunjinSong/PM-Avatars.

Human motion capture from monocular videos has made significant progress in recent years. However, modern approaches often produce temporal artifacts, e.g. in form of jittery motion and struggle to achieve smooth and physically plausible motions. Explicitly integrating physics, in form of internal forces and exterior torques, helps alleviating these artifacts. Current state-of-the-art approaches make use of an automatic PD controller to predict torques and reaction forces in order to re-simulate the input kinematics, i.e. the joint angles of a predefined skeleton. However, due to imperfect physical models, these methods often require simplifying assumptions and extensive preprocessing of the input kinematics to achieve good performance. To this end, we propose a novel method to selectively incorporate the physics models with the kinematics observations in an online setting, inspired by a neural Kalman-filtering approach. We develop a control loop as a meta-PD controller to predict internal joint torques and external reaction forces, followed by a physics-based motion simulation. A recurrent neural network is introduced to realize a Kalman filter that attentively balances the kinematics input and simulated motion, resulting in an optimal-state dynamics prediction. We show that this filtering step is crucial to provide an online supervision that helps balancing the shortcoming of the respective input motions, thus being important for not only capturing accurate global motion trajectories but also producing physically plausible human poses. The proposed approach excels in the physics-based human pose estimation task and demonstrates the physical plausibility of the predictive dynamics, compared to state of the art. The code is available on https://github.com/cuongle1206/OSDCap.

Scene flow estimation is an essential ingredient for a variety of real-world applications, especially for autonomous agents, such as self-driving cars and robots. While recent scene flow estimation approaches achieve reasonable accuracy, their applicability to real-world systems additionally benefits from a reliability measure. Aiming at improving accuracy while additionally providing an estimate for uncertainty, we propose DiffSF that combines transformer-based scene flow estimation with denoising diffusion models. In the diffusion process, the ground truth scene flow vector field is gradually perturbed by adding Gaussian noise. In the reverse process, starting from randomly sampled Gaussian noise, the scene flow vector field prediction is recovered by conditioning on a source and a target point cloud. We show that the diffusion process greatly increases the robustness of predictions compared to prior approaches resulting in state-of-the-art performance on standard scene flow estimation benchmarks. Moreover, by sampling multiple times with different initial states, the denoising process predicts multiple hypotheses, which enables measuring the output uncertainty, allowing our approach to detect a majority of the inaccurate predictions. The code is available at https://github.com/ZhangYushan3/DiffSF.

During the operation of industrial robots, unusual events may endanger the safety of humans and the quality of production. When collecting data to detect such cases, it is not ensured that data from all potentially occurring errors is included as unforeseeable events may happen over time. Therefore, anomaly detection (AD) delivers a practical solution, using only normal data to learn to detect unusual events. We introduce a dataset that allows training and benchmarking of anomaly detection methods for robotic applications based on machine data which will be made publicly available to the research community. As a typical robot task the dataset includes a pick-and-place application which involves movement, actions of the end effector, and interactions with the objects of the environment. Since several of the contained anomalies are not task-specific but general, evaluations on our dataset are transferable to other robotics applications as well. In addition, we present multivariate time-series flow (MVT-Flow) as a new baseline method for anomaly detection: It relies on deep-learning-based density estimation with normalizing flows, tailored to the data domain by taking its structure into account for the architecture. Our evaluation shows that MVT-Flow outperforms baselines from previous work by a large margin of 6.2% in area under receiving operator characteristic.

Traditionally, monocular 3D human pose estimation employs a machine learning model to predict the most likely 3D pose for a given input image. However, a single image can be highly ambiguous and induces multiple plausible solutions for the 2D-3D lifting step, which results in overly confident 3D pose predictors. To this end, we propose DiffPose, a conditional diffusion model that predicts multiple hypotheses for a given input image. Compared to similar approaches, our diffusion model is straightforward and avoids intensive hyperparameter tuning, complex network structures, mode collapse, and unstable training. Moreover, we tackle the problem of over-simplification of the intermediate representation of the common two-step approaches which first estimate a distribution of 2D joint locations via joint-wise heatmaps and consecutively use their maximum argument for the 3D pose estimation step. Since such a simplification of the heatmaps removes valid information about possibly correct, though labeled unlikely, joint locations, we propose to represent the heatmaps as a set of 2D joint candidate samples. To extract information about the original distribution from these samples, we introduce our embedding transformer which conditions the diffusion model. Experimentally, we show that DiffPose improves upon the state of the art for multi-hypothesis pose estimation by 3-5% for simple poses and outperforms it by a large margin for highly ambiguous poses.1

We tackle the task of scene flow estimation from point clouds. Given a source and a target point cloud, the objective is to estimate a translation from each point in the source point cloud to the target, resulting in a 3D motion vector field. Previous dominant scene flow estimation methods require complicated coarse-to-fine or recurrent architectures as a multi-stage refinement. In contrast, we propose a significantly simpler single-scale one-shot global matching to address the problem. Our key finding is that reliable feature similarity between point pairs is essential and sufficient to estimate accurate scene flow. We thus propose to decompose the feature extraction step via a hybrid local-global-cross transformer architecture which is crucial to accurate and robust feature representations. Extensive experiments show that the proposed Global Matching Scene Flow (GMSF) sets a new state-of-the-art on multiple scene flow estimation benchmarks. On FlyingThings3D, with the presence of occlusion points, GMSF reduces the outlier percentage from the previous best performance of 27.4% to 5.6%. On KITTI Scene Flow, without any fine-tuning, our proposed method shows state-of-the-art performance. On the Waymo-Open dataset, the proposed method outperforms previous methods by a large margin. The code is available at https://github.com/ZhangYushan3/GMSF.

Industrial defect detection is commonly addressed with anomaly detection (AD) methods where no or only incomplete data of potentially occurring defects is available. This work discovers previously unknown problems of student-teacher approaches for AD and proposes a solution, where two neural networks are trained to produce the same output for the defect-free training examples. The core assumption of student-teacher networks is that the distance between the outputs of both networks is larger for anomalies since they are absent in training. However, previous methods suffer from the similarity of student and teacher architecture, such that the distance is undesirably small for anomalies. For this reason, we propose asymmetric student-teacher networks (AST). We train a normalizing flow for density estimation as a teacher and a conventional feed-forward network as a student to trigger large distances for anomalies: The bijectivity of the normalizing flow enforces a divergence of teacher outputs for anomalies compared to normal data. Outside the training distribution the student cannot imitate this divergence due to its fundamentally different architecture. Our AST network compensates for wrongly estimated likelihoods by a normalizing flow, which was alternatively used for anomaly detection in previous work. We show that our method produces state-of-the-art results on the two currently most relevant defect detection datasets MVTec AD and MVTec 3D-AD regarding image-level anomaly detection on RGB and 3D data.

A long-standing goal in the field of sensory substitution is enabling sound perception for deaf and hard of hearing (DHH) people by visualizing audio content. Different from existing models that translate to hand sign language, between speech and text, or text and images, we target immediate and low-level audio to video translation that applies to generic environment sounds as well as human speech. Since such a substitution is artificial, with-out labels for supervised learning, our core contribution is to build a mapping from audio to video that learns from unpaired examples via high-level constraints. For speech, we additionally disentangle content from style, such as gender and dialect. Qualitative and quantitative results, including a human study, demonstrate that our unpaired translation approach maintains important audio features in the generated video and that videos of faces and numbers are well suited for visualizing high-dimensional audio features that can be parsed by humans to match and distinguish between sounds and words. Project website: https://chunjinsong.github.io/audioviewer

This paper addresses the problem of cross-dataset generalization of 3D human pose estimation models. Testing a pre-trained 3D pose estimator on a new dataset results in a major performance drop. Previous methods have mainly addressed this problem by improving the diversity of the training data. We argue that diversity alone is not sufficient and that the characteristics of the training data need to be adapted to those of the new dataset such as camera view-point, position, human actions, and body size. To this end, we propose AdaptPose, an end-to-end framework that generates synthetic 3D human motions from a source dataset and uses them to fine-tune a 3D pose estimator. AdaptPose follows an adversarial training scheme. From a source 3D pose the generator generates a sequence of 3D poses and a camera orientation that is used to project the generated poses to a novel view. Without any 3D labels or camera information AdaptPose successfully learns to create synthetic 3D poses from the target dataset while only being trained on 2D poses. In experiments on the Human3.6M, MPI-INF-3DHp, 3DPW, and Ski-Pose datasets our method outperforms previous work in cross-dataset evaluations by 14% and previous semi-supervised learning methods that use partial 3D annotations by 16%.

Segmenting an image into its parts is a common pre-process for high-level vision tasks such as image editing. However, annotating masks for supervised training is expensive. Weakly-supervised and unsupervised methods exist, but they depend on the comparison of pairs of images, such as from multi-views, frames of videos, and image augmentation, which limit their applicability. To address this, we propose a GAN-based approach that generates images conditioned on latent masks, thereby alleviating full or weak annotations required by previous approaches. We show that such mask-conditioned image generation can be learned faithfully when conditioning the masks in a hierarchical manner on 2D latent points that define the position of parts explicitly. Without requiring supervision of masks or points, this strategy increases robustness of mask to viewpoint and object position changes. It also lets us generate image-mask pairs for training a segmentation network, which outperforms state-of-the-art unsupervised segmentation methods on established benchmarks. Code can be found at https://github.com/xingzhehe/GANSeg.

Human pose estimation from single images is a challenging problem that is typically solved by supervised learning. Unfortunately, labeled training data does not yet exist for many human activities since 3D annotation requires dedicated motion capture systems. Therefore, we propose an unsupervised approach that learns to predict a 3D human pose from a single image while only being trained with 2D pose data, which can be crowd-sourced and is already widely available. To this end, we estimate the 3D pose that is most likely over random projections, with the likelihood estimated using normalizing flows on 2D poses. While previous work requires strong priors on camera rotations in the training data set, we learn the distribution of camera angles which significantly improves the performance. Another part of our contribution is to stabilize training with normalizing flows on high-dimensional 3D pose data by first projecting the 2D poses to a linear subspace. We outperform the state-of-the-art unsupervised human pose estimation methods on the benchmark datasets Human3.6M and MPI-INF-3DHP in many metrics.

Structured representations such as keypoints are widely used in pose transfer, conditional image generation, animation, and 3D reconstruction. However, their supervised learning requires expensive annotation for each target domain. We propose a self-supervised method that learns to disentangle object structure from the appearance with a graph of 2D keypoints linked by straight edges. Both the keypoint location and their pairwise edge weights are learned, given only a collection of images depicting the same object class. The resulting graph is interpretable, for example, AutoLink recovers the human skeleton topology when applied to images showing people. Our key ingredients are i) an encoder that predicts keypoint locations in an input image, ii) a shared graph as a latent variable that links the same pairs of keypoints in every image, iii) an intermediate edge map that combines the latent graph edge weights and keypoint locations in a soft, differentiable manner, and iv) an inpainting objective on randomly masked images. Although simpler, AutoLink outperforms existing self-supervised methods on the established keypoint and pose estimation benchmarks and paves the way for structure-conditioned generative models on more diverse datasets. Project website: https://xingzhehe.github.io/autolink/.

In industrial manufacturing processes, errors frequently occur at unpredictable times and in unknown manifestations. We tackle the problem of automatic defect detection without requiring any image samples of defective parts. Recent works model the distribution of defect-free image data, using either strong statistical priors or overly simplified data representations. In contrast, our approach handles fine-grained representations incorporating the global and local image context while flexibly estimating the density. To this end, we propose a novel fully convolutional cross-scale normalizing flow (CS-Flow) that jointly processes multiple feature maps of different scales. Using normalizing flows to assign meaningful likelihoods to input samples allows for efficient defect detection on image-level. Moreover, due to the preserved spatial arrangement the latent space of the normalizing flow is interpretable which enables to localize defective regions in the image. Our work sets a new state-of-the-art in image-level defect detection on the benchmark datasets Magnetic Tile Defects and MVTec AD showing a 100% AUROC on 4 out of 15 classes.

Human pose estimation from single images is a challenging problem in computer vision that requires large amounts of labeled training data to be solved accurately. Unfortunately, for many human activities (e.g. outdoor sports) such training data does not exist and is hard or even impossible to acquire with traditional motion capture systems. We propose a self-supervised approach that learns a single image 3D pose estimator from unlabeled multi-view data. To this end, we exploit multi-view consistency constraints to disentangle the observed 2D pose into the underlying 3D pose and camera rotation. In contrast to most existing methods, we do not require calibrated cameras and can therefore learn from moving cameras. Nevertheless, in the case of a static camera setup, we present an optional extension to include constant relative camera rotations over multiple views into our framework. Key to the success are new, unbiased reconstruction objectives that mix information across views and training samples. The proposed approach is evaluated on two benchmark datasets (Human3.6M and MPII-INF-3DHP) and on the in-the-wild SkiPose dataset.

3D human pose estimation from monocular images is a highly ill-posed problem due to depth ambiguities and occlusions. Nonetheless, most existing works ignore these ambiguities and only estimate a single solution. In contrast, we generate a diverse set of hypotheses that represents the full posterior distribution of feasible 3D poses. To this end, we propose a normalizing flow based method that exploits the deterministic 3D-to-2D mapping to solve the ambiguous inverse 2D-to-3D problem. Additionally, uncertain detections and occlusions are effectively modeled by incorporating uncertainty information of the 2D detector as condition. Further keys to success are a learned 3D pose prior and a generalization of the best-of-M loss. We evaluate our approach on the two benchmark datasets Human3.6M and MPI-INF-3DHP, outperforming all comparable methods in most metrics. The implementation is available on GitHub1.

The detection of manufacturing errors is crucial in fabrication processes to ensure product quality and safety standards. Since many defects occur very rarely and their characteristics are mostly unknown a priori, their detection is still an open research question. To this end, we propose DifferNet: It leverages the descriptiveness of features extracted by convolutional neural networks to estimate their density using normalizing flows. Normalizing flows are well-suited to deal with low dimensional data distributions. However, they struggle with the high dimensionality of images. Therefore, we employ a multi-scale feature extractor which enables the normalizing flow to assign meaningful likelihoods to the images. Based on these likelihoods we develop a scoring function that indicates defects. Moreover, propagating the score back to the image enables pixel-wise localization. To achieve a high robustness and performance we exploit multiple transformations in training and evaluation. In contrast to most other methods, ours does not require a large number of training samples and performs well with as low as 16 images. We demonstrate the superior performance over existing approaches on the challenging and newly proposed MVTec AD [4] and Magnetic Tile Defects [14] datasets.

This paper addresses the problem of 3D human pose estimation from single images. While for a long time human skeletons were parameterized and fitted to the observation by satisfying a reprojection error, nowadays researchers directly use neural networks to infer the 3D pose from the observations. However, most of these approaches ignore the fact that a reprojection constraint has to be satisfied and are sensitive to overfitting. We tackle the overfitting problem by ignoring 2D to 3D correspondences. This efficiently avoids a simple memorization of the training data and allows for a weakly supervised training. One part of the proposed reprojection network (RepNet) learns a mapping from a distribution of 2D poses to a distribution of 3D poses using an adversarial training approach. Another part of the network estimates the camera. This allows for the definition of a network layer that performs the reprojection of the estimated 3D pose back to 2D which results in a reprojection loss function. Our experiments show that RepNet generalizes well to unknown data and outperforms state-of-the-art methods when applied to unseen data. Moreover, our implementation runs in real-time on a standard desktop PC.

Roads constitute a major part of the lives of everybody. Heavy use, for instance by cars and especially trucks, and even soil movement lead to visible damages. While major roads are regularly inspected, smaller roads often lack attention. It is therefore of great interest to have camera-based systems which can automatically detect and even classify damages.This report presents a system developed by the authors as part of the Road Damage Detection and Classification Challenge at the 2018 IEEE Big Data Cup [1]. Further contributions made here are techniques to augment the small set of training data. As a major contribution we also propose refinements to the dataset and evaluation metric to improve the challenge.

Modern video coding standards like High Efficiency Video Coding (HEVC) provide superior coding efficiency. However, this does not state true for complex and hard to predict textures which require high bit rates to achieve a high quality. To overcome this limitation of HEVC, texture synthesis frameworks were proposed in previous works. However, these frameworks only result in good reconstruction quality if the decomposition into synthesizable and non-synthesizable regions is either known or trivial. The frameworks fail for more challenging content, e.g. for content with fine non-synthesizable details within synthesizable regions. To enable texture synthesis-based video coding with high quality for this content, we propose sophisticated detail-aware decomposition techniques in this paper. These techniques are based on an initial coarse segmentation step followed by a refinement step that detects even small differences in the previously segmented region. With this new approach, we are able to achieve average luma BD-rate gains of 13.77% over HEVC and 3.03% over the closest related work from the literature. Furthermore, the considerably improved visual quality in addition to the bit rate savings is confirmed by comprehensive subjective tests.

In recent years, there has been a tremendous improvement in video coding algorithms. This improvement resulted in 2013 in the standardization of the first version of High Efficiency Video Coding (HEVC) which now forms the state-of-theart with superior coding efficiency. Nevertheless, the development of video coding algorithms did not stop as HEVC still has its limitations. Especially for complex textures HEVC reveals one of its limitations. As these textures are hard to predict, very high bit rates are required to achieve a high quality. Texture synthesis was proposed as solution for this limitation in previous works. However, previous texture synthesis frameworks only prevailed if the decomposition into synthesizable and non-synthesizable regions was either known or very easy. In this paper, we address this scenario with a texture synthesis framework based on detail-aware image decomposition techniques. Our techniques are based on a multiple-steps coarse-to-fine approach in which an initial decomposition is refined with awareness for small details. The efficiency of our approach is evaluated objectively and subjectively: BD-rate gains of up to 28.81% over HEVC and up to 12.75% over the closest related work were achieved. Our subjective tests indicate an improved visual quality in addition to the bit rate savings.

The High Efficiency Video Coding (HEVC) standard provides superior coding efficiency compared to its predecessors. Nevertheless, the encoding of complex and thus hardly to predict textures either requires high bit rates or results in low quality of the reconstructed signal. To compensate for this limitation of HEVC, we propose a sophisticated texture synthesis framework which solves multiple lacks of previous texture synthesis approaches. By easing the bit rate cost for synthesizable regions and reallocating the freed bit rate resources to non-synthesizable regions, for high-value soccer content we are able to achieve average BD-rate gains of 21.9% for all-intra, 17.6% for low delay, and 16.3% for random access, respectively, while maintaining the same objective quality for the latter. Subjective tests for the synthesizable regions confirm the objectively measured convincing results. The general applicability of our method is confirmed for other types of content.

This article tackles the problem of estimating non-rigid human 3D shape and motion from image sequences taken by uncalibrated cameras. Similar to other state-of-the-art solutions we factorize 2D observations in camera parameters, base poses and mixing coefficients. Existing methods require sufficient camera motion during the sequence to achieve a correct 3D reconstruction. To obtain convincing 3D reconstructions from arbitrary camera motion, our method is based on a-priorly trained base poses. We show that strong periodic assumptions on the coefficients can be used to define an efficient and accurate algorithm for estimating periodic motion such as walking patterns. For the extension to non-periodic motion we propose a novel regularization term based on temporal bone length constancy. In contrast to other works, the proposed method does not use a predefined skeleton or anthropometric constraints and can handle arbitrary camera motion. We achieve convincing 3D reconstructions, even under the influence of noise and occlusions. Multiple experiments based on a 3D error metric demonstrate the stability of the proposed method. Compared to other state-of-the-art methods our algorithm shows a significant improvement.