Weakly-supervised pre-training for 3D human pose estimation via perspective knowledge

3D human pose estimation aims to reconstruct the human skeleton of all the individuals in a scene by detecting several body joints. The creation of accurate and efficient methods is required for several real-world applications including animation, human–robot interaction, surveillance systems or sports, among many others. However, several obstacles such as occlusions, random camera perspectives, or the scarcity of 3D labelled data, have been hampering the models’ performance and limiting their deployment in real-world scenarios. The higher availability of cameras has led researchers to explore multi-view solutions due to the advantage of being able to exploit different perspectives to reconstruct the pose. Most existing reviews focus mainly on monocular 3D human pose estimation and a comprehensive survey only on multi-view approaches to determine the 3D pose has been missing since 2012. Thus, the goal of this survey is to fill that gap and present an overview of the methodologies related to 3D pose estimation in multi-view settings, understand what were the strategies found to address the various challenges and also, identify their limitations. According to the reviewed articles, it was possible to find that most methods are fully-supervised approaches based on geometric constraints. Nonetheless, most of the methods suffer from 2D pose mismatches, to which the incorporation of temporal consistency and depth information have been suggested to reduce the impact of this limitation, besides working directly with 3D features can completely surpass this problem but at the expense of higher computational complexity. Models with lower supervision levels were identified to overcome some of the issues related to 3D pose, particularly the scarcity of labelled datasets. Therefore, no method is yet capable of solving all the challenges associated with the reconstruction of the 3D pose. Due to the existing trade-off between complexity and performance, the best method depends on the application scenario. Therefore, further research is still required to develop an approach capable of quickly inferring a highly accurate 3D pose with bearable computation cost. To this goal, techniques such as active learning, methods that learn with a low level of supervision, the incorporation of temporal consistency, view selection, estimation of depth information and multi-modal approaches might be interesting strategies to keep in mind when developing a new methodology to solve this task. • First review only on multi-view, multi-modal methods to estimate 3D pose since 2012. • Multi-view allows capturing the full body geometry, making 3D pose estimation easier. • Real-world applications include sports, broadcasting, rehabilitation or animation. • Finding a fast, accurate method with low computational cost remains a challenge. • Multi-modal methods or view selection can lead to an efficient and effective model.

Recently, the emergence of Kinect systems has demonstrated the benefits of predicting an intermediate body part labeling for 3D human pose estimation, in conjunction with RGB-D imagery. The availability of depth information plays a critical role, so an important question is whether a similar representation can be developed with sufficient robustness in order to estimate 3D pose from RGB images. This paper provides evidence for a positive answer, by leveraging (a) 2D human body part labeling in images, (b) second-order label-sensitive pooling over dynamically computed regions resulting from a hierarchical decomposition of the body, and (c) iterative structured-output modeling to contextualize the process based on 3D pose estimates. For robustness and generalization, we take advantage of a recent large-scale 3D human motion capture dataset, Human3.6M[18] that also has human body part labeling annotations available with images. We provide extensive experimental studies where alternative intermediate representations are compared and report a substantial 33% error reduction over competitive discriminative baselines that regress 3D human pose against global HOG features.

In the field of 3D Human Pose Estimation and Reconstruction based on body joints extracted from a 2D image; Exists challenges like self-occlusion and depth perception. These problems hinder approximate estimations. This article proposes a hybrid method that consists of semantic segmentation, sparse representation and 2D pose estimation for 3D estimation. Specifically, we train two fully Convolution neural networks (FCNs) to estimate 2D pose and semantic segmentation. Then, we narrow the estimation down using basic human body structure and results from the FCNs. Subsequently, utilizing the estimated 2D pose of the earlier step and a sparse representation model the 3D pose will estimate. Using a Convolution Neural Networks learning method for 2D human pose estimation and then by sparse representation will estimate. Results of the proposed method, show improvement relative to the previous works on 3D Human pose estimation. 3D Human Pose Estimation by this method demonstrates a significant mean error reduction compared to earlier studies.

Deep 3D human pose estimation: A review

Generalization is very important for pose estimation, especially for 3D pose estimation where small changes in the 2D images could trigger structural changes in the 3D space. To achieve generalization, the system needs to have the capability of detecting estimation errors by double-checking the projection coherence between the 3D and 2D spaces and adapting its network inference process based on this feedback. Current pose estimation is one-time feed-forward and lacks the capability to gather feedback and adapt the inference outcome. To address this problem, we propose to explore the concept of progressive inference where the network learns an observer to continuously detect the prediction error based on constraints matching, as well as an adjuster to refine its inference outcome based on these constraints errors. Within the context of 3D hand pose estimation, we find that this observer-adjuster design is relatively unstable since the observer is operating in the 2D image domain while the adjuster is operating in the 3D domain. To address this issue, we propose to construct two sets of observers-adjusters with complementary constraints from different perspectives. They operate in a dynamic sequential manner controlled by a decision network to progressively improve the 3D pose estimation. We refer to this method as Cross-Constrained Progressive Inference (CCPI). Our extensive experimental results on FreiHAND and HO-3D benchmark datasets demonstrate that the proposed CCPI method is able to significantly improve the generalization capability and performance of 3D hand pose estimation.

In this paper, we propose 3DBodyTex.Pose, a dataset that addresses the task of 3D human pose estimation in-the-wild. Generalization to in-the-wild images remains limited due to the lack of adequate datasets. Existent ones are usually collected in indoor controlled environments where motion capture systems are used to obtain the 3D ground-truth annotations of humans. 3DBodyTex.Pose offers high quality and rich data containing 405 different real subjects in various clothing and poses, and 81k image samples with ground-truth 2D and 3D pose annotations. These images are generated from 200 viewpoints among which 70 challenging extreme viewpoints. This data was created starting from high resolution textured 3D body scans and by incorporating various realistic backgrounds. Retraining a state-of-the-art 3D pose estimation approach using data augmented with 3DBodyTex.Pose showed promising improvement in the overall performance, and a sensible decrease in the per joint position error when testing on challenging viewpoints. The 3DBodyTex.Pose is expected to offer the research community with new possibilities for generalizing 3D pose estimation from monocular in-the-wild images.

The neural network based approach for 3D human pose estimation from monocular images has attracted growing interest. However, annotating 3D poses is a labor-intensive and expensive process. In this paper, we propose a novel self-supervised approach to avoid the need of manual annotations. Different from existing weakly/self-supervised methods that require extra unpaired 3D ground-truth data to alleviate the depth ambiguity problem, our method trains the network only relying on geometric knowledge without any additional 3D pose annotations. The proposed method follows the two-stage pipeline: 2D pose estimation and 2D-to-3D pose lifting. We design the transform re-projection loss that is an effective way to explore multi-view consistency for training the 2D-to-3D lifting network. Besides, we adopt the confidences of 2D joints to integrate losses from different views to alleviate the influence of noises caused by the self-occlusion problem. Finally, we design a two-branch training architecture, which helps to preserve the scale information of re-projected 2D poses during training, resulting in accurate 3D pose predictions. We demonstrate the effectiveness of our method on two popular 3D human pose datasets, Human3.6M and MPI-INF-3DHP. The results show that our method significantly outperforms recent weakly/self-supervised approaches.

We present a deployment friendly, fast bottom-up framework for multi-person 3D human pose estimation. We adopt a novel neural representation of multi-person 3D pose which unifies the position of person instances with their corresponding 3D pose representation. This is realized by learning a generative pose embedding which not only ensures plausible 3D pose predictions, but also eliminates the usual keypoint grouping operation as employed in prior bottom-up approaches. Further, we propose a practical deployment paradigm where paired 2D or 3D pose annotations are unavailable. In the absence of any paired supervision, we leverage a frozen network, as a teacher model, which is trained on an auxiliary task of multi-person 2D pose estimation. We cast the learning as a cross-modal alignment problem and propose training objectives to realize a shared latent space between two diverse modalities. We aim to enhance the model’s ability to perform beyond the limiting teacher network by enriching the latent-to-3D pose mapping using artificially synthesized multi-person 3D scene samples. Our approach not only generalizes to in-the-wild images, but also yields a superior trade-off between speed and performance, compared to prior top-down approaches. Our approach also yields state-of-the-art multi-person 3D pose estimation performance among the bottom-up approaches under consistent supervision levels.

The estimation of 3D body pose and shape has always been a challenging problem due to various reasons, such as the ambiguity in 2D images and complex articulated structure of the human body. In order to solve the ill-conditioned problems, in this paper, we bring up an end-to-end method to estimate 3D human shape and pose from multi-view RGB images. In the proposed framework, we first implement a CNN embedded with attention module to extract the image feature and design the view-pooling layer to combine the features from multiple views. Then we adopt a regression network with a novel geometric constraint of body limbs to estimate 3D human pose and shape. Additionally, during the training process, we employ the idea of adversarial learning in our model to help regress accurate pose and shape parameters. Extensive experiments are conducted on Human3.6M and MPI-INF-3DHP datasets, and our method achieves competitive results in the 3D pose and shape estimation task.

It is widely shown that Convolutional Neural Networks (CNNs) are vulnerable to adversarial examples on most recognition tasks, such as image classification and segmentation. However, few work studies the more complicated task – 3D human pose estimation. This task often requires large-scale datasets, specialized network architectures, and it can be solved either from single-view RGB images or from multi-view RGB images. In this paper, we make the first attempt to investigate the robustness of current state-of-the-art 3D human pose estimation methods. To this end, we build four representative baseline models, where most of the current methods can be generally classified as one of them. Furthermore, we design targeted adversarial attacks to detect whether 3D pose estimators are robust to different camera parameters. For different types of methods, we present a comprehensive study of their robustness on the large-scale Human3.6M benchmark. Our work shows that different methods vary significantly in their resistance to adversarial attacks. Through extensive experiments, we show that multi-view 3D pose estimators can be more vulnerable to adversarial examples. We believe that our efforts can shed light on future works to design more robust 3D human pose estimators.

While monocular 3D pose estimation seems to have achieved very accurate results on the public datasets, their generalization ability is largely overlooked. In this work, we perform a systematic evaluation of the existing methods and find that they get notably larger errors when tested on different cameras, human poses and appearance. To address the problem, we introduce VirtualPose, a two-stage learning framework to exploit the hidden “free lunch” specific to this task, i.e.generating infinite number of poses and cameras for training models at no cost. To that end, the first stage transforms images to abstract geometry representations (AGR), and then the second maps them to 3D poses. It addresses the generalization issue from two aspects: (1) the first stage can be trained on diverse 2D datasets to reduce the risk of over-fitting to limited appearance; (2) the second stage can be trained on diverse AGR synthesized from a large number of virtual cameras and poses. It outperforms the SOTA methods without using any paired images and 3D poses from the benchmarks, which paves the way for practical applications. Code is available at https://github.com/wkom/VirtualPose. KeywordAbsolute 3D human pose estimation

Automatically estimating 3D human poses in video and inferring their meanings play an essential role in many human-centered automation systems. Existing researches made remarkable progresses by first estimating 2D human joints in video and then reconstructing 3D human pose from the 2D joints. However, mono-directionally reconstructing 3D pose from 2D joints ignores the interaction between information in 3D space and 2D space, losses rich information of original video, therefore limits the ceiling of estimation accuracy. To this end, this paper proposes a bidirectional 2D-3D transformation framework that bidirectionally exchanges 2D and 3D information and utilizes video information to estimate an offset for refining 3D human pose. In addition, a bone-length stability loss is utilized for the purpose of exploring human body structure to make the estimated 3D pose more natural and to further increase the overall accuracy. By evaluation, estimation error of the proposed method, measured by the mean per joint position error (MPJPE), is only 46.5 mm, which is much lower than state-of-the-art methods under the same experimental condition. The improvement on accuracy will make machines to better understand human poses for building superior human-centered automation systems. <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Note to Practitioners</i> —This paper was motivated by the demand of human-centered automation systems needing to accurately understand human poses. Existing approaches mainly focus on inferring 3D human pose from 2D joints mono-directionally. Although they made remarkable contributions to estimating 3D human pose in such a mono-directional way, we found that they ignore the 2D-3D interaction and do not use original video when inferring 3D pose from 2D joints. This paper therefore suggests a bidirectional 2D-3D transformation that exchanges 2D and 3D information and utilizes video information to estimate more accurate 3D human pose for human-centered automation systems. This work is a pioneering attempt of interactively using 2D and 3D information for more accurate estimation of human pose. Benefited from the state-of-the-art accuracy, the proposed approach is expected to make significant contributions to many human-centered automation systems, such as human-machine interaction, biomimetic manipulation, and automatic surveillance systems.

Enhancement of human 3D pose estimation using a novel concept of depth prediction with pose alignment from a single 2D image

Face pose information is valuable for a variety of applications including unconstrained face recognition, natural human computer interfaces, and video database indexing. 3D pose estimation is a critical requirement for accurate face recognition using view varying representations, such as 2D intensity images. 3D pose extraction in this context requires 3D information, which is present in image sequences if we assume the moving objects in the sequence are primarily rigid. This paper presents a motion based pose estimation system which computes the 3D pose of the head in each frame of a video sequence. Generic low level features, such as corners, are identified and tracked in the video stream. The feature tracks are processed by a shape from motion algorithm which produces estimates of 3D geometry and pose. The geometry and pose estimates are considered together with facial structure constraints, temporal constraints, and initial pose estimates to refine knowledge of the specific face structure and its pose. We describe this system and show that it works on human faces. This is significant because the face has many smooth surfaces which make it difficult to extract dense intensity features.

ESMformer: Error-aware self-supervised transformer for multi-view 3D human pose estimation