PyTorch implementation of A simple yet effective baseline for 3d human pose estimation [Martinez+, ICCV'17].
Todo:
- Provide trained weight
- Provide tutorials to predict 3D pose from 2D pose input
- Train models on Stacked Hourglass output
MPJPE [mm]:
Avg | Direct | Discuss | Eating | Greet | Phone | Photo | Pose | Purch | Sitting | SittingD | Smoke | Wait | WalkD | Walk | WalkT | |
---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Paper | 45.5 | 37.7 | 44.4 | 40.3 | 42.1 | 48.2 | 54.9 | 44.4 | 42.1 | 54.6 | 58.0 | 45.1 | 46.4 | 47.6 | 36.4 | 40.4 |
This repo | 43.3 | 35.7 | 41.6 | 40.1 | 40.4 | 45.0 | 52.0 | 42.9 | 38.0 | 53.2 | 55.4 | 43.5 | 43.3 | 43.3 | 33.7 | 35.6 |
Both were trained on GT 2D pose input from multiple actions.
Get h36m.zip
by following author's repo, place it under dataset
, and unzip it.
$ cd dataset
$ unzip h36m.zip
$ pip install -r requirements.txt
$ ./tools/train.py OUTPUT_DIR ./output
You'll find trained weight and tensorboard event file under ./output
directory.
$ ./tools/test.py OUTPUT_DIR ./output MODEL.WEIGHT ${PATH_TO_WEIGHT}
You'll find evaluation results in a JSON file under ./output
directory.
A simple yet effective baseline for 3d human pose estimation
Julieta Martinez, Rayat Hossain, Javier Romero, James J. Little
[Paper][Author's implementation]
@inproceedings{martinez_2017_3dbaseline,
title={A simple yet effective baseline for 3d human pose estimation},
author={Martinez, Julieta and Hossain, Rayat and Romero, Javier and Little, James J.},
booktitle={ICCV},
year={2017}
}