This is the Tensorflow implementation of our ECCV2020 paper "JGR-P2O: Joint Graph Reasoning based Pixel-to-Offset Prediction Network for 3D Hand Pose Estimation from a Single Depth Image"
The key ideas of JGR-P2O are two-fold: a) explicitly modeling the dependencies among joints and the relations between the pixels and the joints with the joint graph reasoning module for better local feature representation learning; b) unifying the dense pixel-wise offset predictions and direct joint regression for end-to-end training.
Figure 1: Overview of JGR-P2O.
-
Requirements:
- python3, tensorflow 1.12-14
-
datasets:
- Download the NYU, ICVL, and MSRA datasets.
- Download the hand centers files from V2V-PoseNet for data preprocessing.
- Thanks CrossInfoNet for providing a good reference of data preprocessing.
-
training and testing:
- Here we provide an example for NYU training and testing, running on ICVL and MSRA is the same way.
- training: python main_nyu.py --phase train --data_root /home/data/3D_hand_pose_estimation/NYU/
- testing: python main_nyu.py --phase test --data_root /home/data/3D_hand_pose_estimation/NYU/
-
Models:
- The pretrained models on NYU and ICVL are included in the "checkpoint" directory of this repository.