Question about inference

[dongho-Han]

Before actually deploying in the wild, I'm trying to inference the model with dataset you shared. But, I have some problems in running the attached code.

(1) For build_dataset.py, do I have to make .npz file for

(i) each .bvh & .mvnx

• python build_dataset.py --filepath /data/DivaTrack_Dataset/meta_postprocessed/S0_159_exp/S0_001 --dataset_name S0_001

(ii) each exp titles:

• python build_dataset.py --filepath /data/DivaTrack_Dataset/meta_postprocessed/S0_159_exp --dataset_name S0_159_exp

(iii) for whole postprocessed dataset

• python build_dataset.py --filepath /data/DivaTrack_Dataset/meta_postprocessed --dataset_name Divatrack

I thought (ii) is right, but wonder it's true or not.

(2) For main.py, I think there are some missing files. I think the inference code should be like this. python main.py --mode infer --model DTVel --target pose --valid_set S0_159_exp_valid.npz --cp_datetime 0510_063301 --resume_iter 1 However, this error comes out.

FileNotFoundError: [Errno 2] No such file or directory: 'expr/checkpoints/0113_210014_FCCalib_6000.pth'

Because, from this line, https://github.com/robindsyang/DivaTrack/blob/3dd7303421bdfee643619aaf4d9410fe5e73bc8d/model/DivaTrackVel.py#L299

the '/0113_210014_FCCalib_6000.pth' file is not given from this repo. Could you share the file?

(3) Could you share any other datasets for testing this model? from the code, I think other dataset could be used, but it's not mentioned in the repo README.

Thanks!

[robindsyang]

Sorry for the late reply.

Q1. (iii) is correct. All motion clips are packed to a single .npz file.

Q2. You may need to write some code to train the calibration model using FC_calib.py and meta22skel npz files.

Q3. TotalCapture dataset: https://cvssp.org/data/totalcapture/