search

IDEA-Research / HumanTOMATO

[ICML 2024] 🍅HumanTOMATO: Text-aligned Whole-body Motion Generation
https://lhchen.top/HumanTOMATO
Other
240 stars 6 forks source link

OpenTMA: Motion-X checkpoints dimension #17

Open HaroldChen19 opened 1 month ago

HaroldChen19 commented 1 month ago

Awesome work!

But I met a problem:

When I use the motionx checkpoints, Line249 and Line251 in 'OpenTMA/tma/models/modeltype/temos.py', there is a "size mismatch for main.0.weight: copying a param with shape torch.Size([512, 309, 4]) from checkpoint, the shape in current model is torch.Size([512, 619, 4]). Copying a param with shape torch.Size([512, 309, 4]) from checkpoint, the shape in current model is torch.Size([512, 619, 4])."

I don't know if it's a problem with my Motion-X data processing or with the checkpoint.

LinghaoChan commented 1 month ago

@HaroldChen19 Could you please copy your error info, codes, and the motion tensor shape to me? Let me have a check.

HonglinChu commented 2 weeks ago

@HaroldChen Hi，I meet the same error, have your solved it?

HonglinChu commented 2 weeks ago

image when i run "python -m train --cfg configs/configs_temos/MotionX-TMR.yaml --cfg_assets configs/assets.yaml --nodebug"

LinghaoChan commented 2 weeks ago

@HonglinChu @HaroldChen19

When I load the file datasets/Motion-X/motion_data/vector_623/aist/subset_0000/Dance_Break_3_Step_clip_1.npy, the data dimension is (289, 313). What about yours?

HonglinChu commented 2 weeks ago

@LinghaoChan Thanks for your reply, i check the data dimension as follow: image We used the code https://github.com/IDEA-Research/Motion-X/blob/main/tomato_represenation/motion_representation.py to process motionx, and we get 623 dim, i don't know why it is different with yours

LinghaoChan commented 2 weeks ago

@LinghaoChan Thanks for your reply, i check the data dimension as follow: image We used the code https://github.com/IDEA-Research/Motion-X/blob/main/tomato_represenation/motion_representation.py to process motionx, and we get 623 dim, i don't know why it is different with yours

https://github.com/IDEA-Research/Motion-X/blob/896241e5ca3cd46ea964058d476823c0a9fbc55b/tomato_represenation/motion_representation.py#L323

This rotation should be removed. @shunlinlu will act soon.

HonglinChu commented 2 weeks ago

OK, I think the format A is right？ image Another problem，motion-x repository is missing the generative scripts of mean and std. I follow the humanld3d script，but it needs minor changes. So, can you provide the complete modified script? image https://github.com/EricGuo5513/HumanML3D/blob/main/cal_mean_variance.ipynb

LinghaoChan commented 2 weeks ago

yes, I will revise the readme accordingly.