Abstract:Multi-person pose estimation generally follows top-down and bottom-up paradigms. The top-down paradigm detects all human boxes and then performs single-person pose estimation on each ROI. The bottom-up paradigm locates identity-free keypoints and then groups them into individuals. Both of them use an extra stage to build the relationship between human instance and corresponding keypoints (e.g., human detection in a top-down manner or a grouping process in a bottom-up manner). The extra stage leads to a high computation cost and a redundant two-stage pipeline. To address the above issue, we introduce a fine-grained body representation method. Concretely, the human body is divided into several local parts and each part is represented by an adaptive point. The novel body representation is able to sufficiently encode the diverse pose information and effectively model the relationship between human instance and corresponding keypoints in a single-forward pass. With the proposed body representation, we further introduce a compact single-stage multi-person pose regression network, called AdaptivePose++, which is the extended version of AAAI-22 paper AdaptivePose. During inference, our proposed network only needs a single-step decode operation to estimate the multi-person pose without complex post-processes and refinements. Without any bells and whistles, we achieve the most competitive performance on representative 2D pose estimation benchmarks MS COCO and CrowdPose in terms of accuracy and speed. In particular, AdaptivePose++ outperforms the state-of-the-art SWAHR-W48 and CenterGroup-W48 by 3.2 AP and 1.4 AP on COCO mini-val with faster inference speed. Furthermore, the outstanding performance on 3D pose estimation datasets MuCo-3DHP and MuPoTS-3D further demonstrates its effectiveness and generalizability on 3D scenes.

Pose Estimation Based on Pose Cluster and Candidates Recombination

Posterior constraints for double-counting problem in clustered pose estimation

Human Pose Estimation with Multiple Mixture Parts Model Based on Upper Body Categories.

An Effective Solution to Double Counting Problem in Human Pose Estimation

Articulated Pose Estimation with Global Body Orientation

Human Pose Estimation Using Exemplars and Part Based Refinement

Top-Down Meets Bottom-Up for Multi-Person Pose Estimation

2D Human Pose Estimation Using Multi-level Dynamic Model

Robust Estimation of 3D Human Poses from a Single Image

Articulated pose estimation with flexible mixtures-of-parts

Pose estimation and body segmentation based on hierarchical searching tree

Face Pose Estimation Based on Eigenspace Analysis and Fuzzy Clustering

A Compact and Powerful Single-Stage Network for Multi-Person Pose Estimation

Pose and Shape Estimation with Discriminatively Learned Parts

Pose Segmentation and Estimation Based on Pictorial Structure Model

Research on 3D Human Pose Estimation Technique Based on Multi-View Information Fusion

Sparse Pose Regression Via Componentwise Clustering Feature Point Representation

Towards Locality Similarity Preserving to 3D Human Pose Estimation.

Robust 3D Human Pose Estimation from Single Images or Video Sequences

A Hierarchical Pose Estimation Method Based on Graph Model

Multi-Person Full Body Pose Estimation