First, clone this repository recursively: git clone --recurse-submodules https://github.com/segfit/segfit.git Then set up the environment as detailed on the Human3D repository. Finally, run pip install -r requirements.txt.

The input path should point to a directory containing PLY files. As neither Human3D nor SegFit require color information, it is sufficient if the PLYs contain only the vertex coordinates. The SMPL-X ground truth path can be omitted. If provided, it should point to a directory containing PKL files with names corresponding to the PLYs. If a scan contains multiple humans and, therefore, multiple ground truth PKLs exist for the scan, make sure that their filenames follow the convention "*'ply-filename'.pkl".

Simply run with: python segfit.py --input_path=<input_path> --human3d_ckpt=<checkpoint_path> --is_input_z_up=<True/False> --smplx_gt_dir=<path_to_SMPLX_ground_truths The is_input_z_upparameter defaults to True and providing SMPLX ground truth parameters is optional. If using the Hi4D dataset, also add the --is_hi4d flag for correct coordinate system alignment.

Run: python segfit.py --input_path=<repository_path>/example/egobody/scans/ --human3d_ckpt=ckpts/human3d.ckpt --is_input_z_up=False --smplx_gt_dir=<repository_path>/example/egobody/truths/.

The pretrained checkpoint for Human3D is taken from the official Human3D repository.