Please prepare data as instructed in the model training sub-page. The training data need to downloaded from here and placed under "${EvoSkeleton}/data" folder:
${EvoSkeleton}
├── data
├── human3.6M
├── your downloaded files
During data space exploration, a function that evaluates the validity of 3D skeletons is used. This function is parametrized with a model propsosed by Ijaz Akhter in CVPR 2015. You need to download the "constraints" folder from here which contains the model parameters and place them under "${EvoSkeleton}/resources" folder:
${EvoSkeleton}
├── recources
├── constraints
To evolve from a population of 3D skeleton (default to Human 3.6M data), go to "${EvoSkeleton}/tools" folder and run
python evolve.py -generate True
To reproduce the experiments in different settings, you need to specify the choice of initial population. For weakly-supervised experiments, you should only start with subject 1 (S1) data (a subset of H36M training data) as follows
python evolve.py -generate True -WS True -SS "S1"
You can even start with extremly scarce data (e.g., 1 percent of S1 data) as follows
python evolve.py -generate True -WS True -SS "0.01S1"
After finished dataset evolution, you can use the saved file for training to see how dataset evolution might help improve model generalization especially when the initial population is scarce.
@inproceedings{akhter2015pose,
title={Pose-conditioned joint angle limits for 3D human pose reconstruction},
author={Akhter, Ijaz and Black, Michael J},
booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
pages={1446--1455},
year={2015}
}