This program is remodeled by moazzem (munnam77) forking 3d-pose-baseline.

Please check the above URL or README-ArashHosseini.md for details of operation.

• Generates a 3D human model from the human skeleton detected by OpenPose.
• Output joint data when generating 3D human model - By reading joint data with 3d-pose-baseline-multi
• Analyze OpenPose data for multiple people.
  • Try to analysis for only one person.
  • Ver2.00 is now supports multiple people tracing. Please check FCRN-DepthPrediction for details.

Install the following in python3 series

• h5py
• tensorflow 1.0 or later

3D skeleton information is created based on Human3.6M. Download the compressed file from below, decompress it, and place it under data.

H36M Data zip (Dropbox)

Since the original learning data hits the 260 character path limit of Windows, the path was simplified and regenerated. Download the compressed file below, unzip it, and place it under experiments.

Learning data zip (GoogleDrive)

1. Analyze data with Openpose Simple Launch Batch
2. Generate data by depth estimation and person index with Depth Estimation
3. Run [OpenposeTo3D.bat] (OpenposeTo3D.bat)
4. You will be asked Directory path by INDEX, so specify the full path of Path by person index in 2. -`{Video file name} json {Execution date and time} _index {Order from the left of the 0th floor}} -For multiple traces, separate execution is required
5. You will be asked if you want to detailed log. If you want to do so, enter yes -If not specified or no, normal log (each parameter file and 3D animated GIF) -In the case of warn, 3D animation GIF is not generated. -If yes, a detailed log is output, and a debug image is output in addition to the log message (slowly)
6. Start processing

MIT