3d-pose-baseline: A Python repository from yamahigashi

3d-pose-baseline

This is the code for the paper

Julieta Martinez, Rayat Hossain, Javier Romero, James J. Little. A simple yet effective baseline for 3d human pose estimation. In ICCV, 2017. https://arxiv.org/pdf/1705.03098.pdf.

The code in this repository was mostly written by Julieta Martinez, Rayat Hossain and Javier Romero.

We provide a strong baseline for 3d human pose estimation that also sheds light on the challenges of current approaches. Our model is lightweight and we strive to make our code transparent, compact, and easy-to-understand.

Dependencies

h5py
tensorflow 1.0 or later

First of all

Watch our video: https://youtu.be/Hmi3Pd9x1BE
Clone this repository and get the data. We provide the Human3.6M dataset in 3d points, camera parameters to produce ground truth 2d detections, and Stacked Hourglass detections.

git clone https://github.com/una-dinosauria/3d-pose-baseline.git
cd 3d-pose-baseline
mkdir data
cd data
wget https://www.dropbox.com/s/e35qv3n6zlkouki/h36m.zip
unzip h36m.zip
rm h36m.zip
cd ..

Quick demo

For a quick demo, you can train for one epoch and visualize the results. To train, run

python src/predict_3dpose.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 1

This should take about <5 minutes to complete on a GTX 1080, and give you around 75 mm of error on the test set.

Now, to visualize the results, simply run

python src/predict_3dpose.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 1 --sample --load 24371

This will produce a visualization similar to this:

openpose/tf-pose-estimation/keras_Realtime_Multi-Person_Pose_Estimation to 3d-Pose-Baseline

Caffe

setup openpose and use --write_json flag to export Pose Keypoints.

Tensorflow

fork tf-pose-estimation and add --output_json flag to export Pose Keypoints like python run_webcam.py --model=mobilenet_thin --resize=432x368 --camera=0 --output_json /path/to/directory, check diff

Keras

fork keras_Realtime_Multi-Person_Pose_Estimation and use python demo_image.py --image sample_images/p1.jpg for single image or python demo_camera.py for webcam feed. check keypoints diff and webcam diff for more info.
Download Pre-trained model below
simply run

python src/openpose_3dpose_sandbox.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --load 4874200 --pose_estimation_json /path/to/json_directory --write_gif --gif_fps 24 , optional --verbose 3 for debug and for interpolation add --interpolation and use --multiplier.

or for 'Real Time'

python3.5 src/openpose_3dpose_sandbox_realtime.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --load 4874200 --pose_estimation_json /path/to/json_directory

Export to DCC application and build skeleton

use --write_json and --write_images flag to export keypoints and frame image from openpose, image will be used as imageplane inside maya.
run python src/openpose_3dpose_sandbox.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --load 4874200 --pose_estimation_json /path/to/json_directory --write_gif --gif_fps 24 .
for interpolation add --interpolation and use --multiplier 0.5.

3d pose baseline now creates a json file 3d_data.json with x, y, z coordinates inside maya folder

change variables in maya/maya_skeleton.py. set threed_pose_baseline to main 3d-pose-baseline and openpose_images to same path as --write_images (step 1)
open maya and import maya/maya_skeleton.py.

maya_skeleton.py will load the data(3d_data.json) to build a skeleton, parenting joints and setting the predicted animation provided by 3d-pose-baseline.

create a imageplane and use created images inside maya/image_plane/ as sequence.

"real-time" stream, openpose > 3d-pose-baseline > maya (soon)
implemented unity stream, check work of Zhenyu Chen openpose_3d-pose-baseline_unity3d

Result

Training

To train a model with clean 2d detections, run:

python src/predict_3dpose.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise

This corresponds to Table 2, bottom row. Ours (GT detections) (MA)

To train on Stacked Hourglass detections, run

python src/predict_3dpose.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh

This corresponds to Table 2, next-to-last row. Ours (SH detections) (MA)

On a GTX 1080 GPU, this takes <8 ms for forward+backward computation, and <6 ms for forward-only computation per batch of 64.

Pre-trained model

We also provide a model pre-trained on Stacked-Hourglass detections, available through google drive

To test the model, decompress the file at the top level of this project, and call

python src/predict_3dpose.py --camera_frame --residual --batch_norm --dropout 0.5 --max_norm --evaluateActionWise --use_sh --epochs 200 --sample --load 4874200

Citing

If you use our code, please cite our work

@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}
}

License

MIT

yamahigashi/3d-pose-baseline