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.

Multiple-Branches Faster Rcnn For Human Parts Detection And Pose Estimation

Human Pose Estimation Based on Multi-branch Convolutional Neural Network

Optimization of Human Pose Detection Based on Mask RCNN

Modelling Human Body Pose for Action Recognition Using Deep Neural Networks

Multi-Human Pose Estimation by Deep Learning-Based Sequential Approach for Human Keypoint Position and Human Body Detection

Towards High Performance One-Stage Human Pose Estimation.

Fast and Robust Multi-Person 3D Pose Estimation from Multiple Views

Multi-Branch Network for Small Human Pose Estimation

DirectPose: Direct End-to-End Multi-Person Pose Estimation

Simultaneous Face Detection And Head Pose Estimation: A Fast And Unified Framework

View Invariant Human Body Detection and Pose Estimation from Multiple Depth Sensors

Human Pose Estimation Based on Lightweight Multi-Scale Coordinate Attention

Joint Model for Human Body Part Instance Segmentation and DensePose Estimation

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

Multi-Person Pose Estimation Using Bounding Box Constraint and LSTM.

Multi-person Pose Estimation with Object Occlusion Information

Learning Delicate Local Representations for Multi-person Pose Estimation

FCPose: Fully Convolutional Multi-Person Pose Estimation with Dynamic Instance-Aware Convolutions

AdaptivePose: Human Parts As Adaptive Points

Improving Multiperson Pose Estimation by Mask-aware Deep Reinforcement Learning

Multi-Person Pose Estimation with Enhanced Channel-wise and Spatial Information