Issues with in model training

[sinapy]

Hey, I've tried to train the model and I faced some confusion:

1. Download AMASS: Store in a folder "amass_raw"". You can train the model for SMPL/SMPL+H or SMPL+X.
From what I understood, I need to install all of these below and put the folders under "amass_raw"

'vald': ['HumanEva', 'MPI_HDM05', 'SFU', 'MPI_mosh'],
'test': ['Transitions_mocap', 'SSM_synced'],
'train': ['CMU', 'MPI_Limits', 'TotalCapture', 'Eyes_Japan_Dataset', 'KIT', 'BioMotionLab_NTroje' ,'BMLhandball' ,
              'BMLmovi', 'EKUT', 'TCD_handMocap',
              'ACCAD']

And the folder structure would look something like this:

I wish the step had clarified this beforehand

Also it takes a very long time to download all of them, specially that the website only allowed me to download one at a time.

Some of the names has also changed, so I've renamed them to the list above

and BioMotionLab_NTroje is BMLrub

2.1 Sample poses from AMASS:

python data/sample_poses.py --sampled_pose_dir <path_for_samples_amass_poses> --amass_dir <amass_dataset_dir>

I didn't understand what is path_for_samples_amass_poses, and also what's it's difference with amass_dataset_dir
In my code I put both path_for_samples_amass_poses and amass_dataset_dir set to ./amass_raw and it worked with some changes to the code
so the final command for me was

conda run python data/sample_poses.py --sampled_pose_dir ./amass_raw/ --amass_dir ./amass_raw/

The other thing that I think should have been said in this step is changing the mode for sample_poses.py. I think for training, I need to run sample_poses.py for the train dataset as well

PoseNDF/data/sample_poses.py

Line 74 in 2a9e4f1

prepare_vposer_datasets(posendf_data_dir, amass_splits, amass_dir, mode='test')

Because if not, I wouldn't have the .npz files of the first layer folders like I did for SSM_synced.

Are the above necessary or should I use the existing .npz files under the second layer folders?

One bug fix is that, for sample_poses.py, you probably don't want to sample the generated .npz files

PoseNDF/data/sample_poses.py

Lines 15 to 30 in 2a9e4f1

	def prepare_vposer_datasets(vposer_dataset_dir, amass_splits, amass_dir, mode='train'):
	keep_rate = 0.3
	amass_datas = sorted(amass_splits[mode])
	for amass_data in amass_datas:
	ds_dir = os.path.join(amass_dir,amass_data)
	seqs = sorted(os.listdir(ds_dir))
	if not os.path.exists(os.path.join(vposer_dataset_dir, amass_data)):
	os.makedirs(os.path.join(vposer_dataset_dir, amass_data))
	for seq in seqs:
	pose_body = []
	root_orient = []
	if 'LICENSE' in seq:
	continue
	out_path = os.path.join(vposer_dataset_dir, amass_data, seq + '.npz')

In the above code, even the below files would be considered for sampling and it will raise an error

so I think you should add this line before line 30:

if '.npz' in seq:
    continue