motokimura / 3d-pose-baseline-pytorch

A simple baseline for 3d human pose estimation in PyTorch

3d-pose-baseline-pytorch

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

Performance

Protocol #1 (no rigid alignment in post-processing)

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.

Preparation

Human3.6M dataset

Get h36m.zip by following author's repo, place it under dataset, and unzip it.

$ cd dataset
$ unzip h36m.zip

Install dependencies

$ pip install -r requirements.txt

Usage

Train model

$ ./tools/train.py OUTPUT_DIR ./output

You'll find trained weight and tensorboard event file under ./output directory.

Evaluate model

$ ./tools/test.py OUTPUT_DIR ./output MODEL.WEIGHT ${PATH_TO_WEIGHT}

You'll find evaluation results in a JSON file under ./output directory.

Paper

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}
}