MotionGPT icon indicating copy to clipboard operation
MotionGPT copied to clipboard

If I only want to evaluate this motionGPT, do I need to build Humanml3D as well

Open LiJianmo opened this issue 1 year ago • 3 comments

When I run: python demo.py --cfg ./configs/config_h3d_stage3.yaml --example ./demos/t2m.txt to evaluate. I always have the error: FileNotFoundError: [Errno 2] No such file or directory: 'datasets/humanml3d\test.txt'

do I need to build humanml3d dataset according to the instruction from its github

LiJianmo avatar Jan 25 '24 02:01 LiJianmo

May I ask if you have already solved the problem? I would appreciate if you could provide a specific solution!

WILLOSCAR avatar Apr 14 '24 16:04 WILLOSCAR

It seems building the complete humanml3d dataset is necessary. Before I built it, I encountered a ValueError: not enough values to unpack (expected 2, got 0) error. However, it ran smoothly after building it.

LiJiaxing0213 avatar May 23 '24 07:05 LiJiaxing0213

datasets/humanl3d must contain all the data files for the dataloader to work. As a result, you need to go to https://github.com/EricGuo5513/HumanML3D for all data preprocessing to obtain them.

The structure of the datasets directory should look like this:

. └── humanml3d ├── Mean.npy ├── Std.npy ├── all.txt ├── new_joint_vecs ├── new_joints ├── test.txt ├── texts ├── texts.zip ├── tmp ├── train.txt ├── train_val.txt └── val.txt

new_joint_vecs and new_joints must be there to make this work

andyl-flwls avatar Jul 26 '24 22:07 andyl-flwls