Thesis research framework developed by Vladyskav Kalyuzhnyy, Utrecht University
Windows or Linux, Python3.9
pip install -r requirements.txt
General System Control:
general.ipynb
Main notebook that executes all framework stages
AlphaPose:
viz_alphapose_output:general.ipynb debugger variable that saves the visualization AlphaPose output in 'ProcessJSON/data/video_n/Camera_n/viz'
AlphaPose/scripts/demo_inference.py
AlphaPose variable controller and 2D joint regression, yolox and Torchreid model deployment
AlphaPose/alphapose/utils/writer.py
update method stores a dummy detection instance for frames in which no detections were captured
AlphaPose/alphapose/utils/pPose_nms.py
write_json method writes all the AlphaPose output to a JSON file
Detectron2:
Detectron2/process_detectron.py
Python file which contains all the methods required to execute and process the data from Detectron2
Detectron2/Detectron2.ipynb
Stand-alone execution of Detectron2, used for debugging
DMMR:
DMMR/cfg_files
Stores the yaml configuration file which contains all the stored parameters for the stage of camera calibration and 3D human reconstruction
DMMR/data/YOUth
YOUth is a replacement data folder. It should be always present in the data folder. general.ipynb will update this folder with the current video's data
DMMR/main.py
Execute main.py for camera calibration and 3d human reconstruction
DMMR/viz_cameras.py
Execute viz_cameras.py for a sequential mesh visualization. This scrip requiers the 'output' file to be present in the DMMR folder
DMMR/output
output data generated from the execution of the main.py file.
output
|--images: empty file
|--meshes: mesh objects of each individual in each frame
|--reprojection: 2D and 3D keyppint information per each frame
|--results: plk files with fitting information of each frame
[FFmpeg] (https://github.com/FFmpeg/FFmpeg)
AlphaPose
[Detectron2] (https://github.com/facebookresearch/detectron2)
[YOLOx] (https://github.com/Megvii-BaseDetection/YOLOX)
[DMMR] (https://github.com/boycehbz/DMMR)