gast

GAST-Net-3DPoseEstimation

Input

(Video from https://github.com/fabro66/GAST-Net-3DPoseEstimation/blob/master/data/video/baseball.mp4)

Output

Usage

Automatically downloads the onnx and prototxt files on the first run. It is necessary to be connected to the Internet while downloading.

For the sample video,

$ python3 gast.py

If you want to specify the input video, put the video file path after the --input option.
You can use --savepath option to change the name of the output file to save. In addition to .mp4, .gif can be specified as the extension of the save file.

$ python3 gast.py --input VIDEO_PATH --savepath SAVE_VIDEO_PATH

The default mode is Single-person 3D pose estimation.
For Two-person 3D pose estimation, you can specify -np 2 option.

$ python3 gast.py -np 2

The Gast-Net adopt YOLOv3 for human detection.
It use implementation of ailia for YOLOv3. You can use -dn option to use implementation of Pytorch. (Then pytorch module is required.)

$ python3 gast.py -dn

Reference

• A Graph Attention Spatio-temporal Convolutional Networks for 3D Human Pose Estimation in Video (GAST-Net)
• YOLOv3
• SORT
• HRNet
• VideoPose3D

Framework

Pytorch

Model Format

ONNX opset=11

Netron

27_frame_17_joint_model.onnx.prototxt
pose_hrnet_w48_384x288.onnx.prototxt
yolov3.opt.onnx.prototxt