search

sail-sg / sewformer

140 stars 14 forks source link

Alignment between garment and clothing #10

Closed sunshineatnoon closed 2 months ago

sunshineatnoon commented 8 months ago

Hi, Thanks for open-sourcing this awesome work.

I tried to "dress" a SMPLH mesh with the clothing from the SewFactory dataset. Specifically, I use fbx_anime.py to produce a SMPLH body with the pose parameter from static__body_info.json in the dataset. Then I visualize the produced SMPLH body and the clothing (e.g., skirt_8_panels_SFQLFR9GJI_skirt_8_panels_SFQLFR9GJI.obj from the tee_1A4AUUJT75_skirt_8_panels_SFQLFR9GJI/static folder) side by side, but it seems they are not aligned (shown below). Is there anything I'm missing here? I suspect there is also a translation for the clothing, if so, where can I find it? Thanks!

Screen Shot 2024-01-19 at 2 02 49 PM
boyang-yu commented 3 months ago

Hi,

Did you figure out why there is such displacement? I just have the same problem here. Besides, when I try to use the pose parameters in the "0000_body_info.json" located in the "poses" folder(e.g. tee_ZZHFVSA17Q_pants_straight_sides_ULX5THI3A5), it does not even return a valid SMPL-H body mesh...

boyang-yu commented 3 months ago

twisted

SMY19999 commented 2 months ago

For the alignment, I just find some code in dataloader, it seems that in /static/spec_config.json, delta is the translation. In code, data/dataset.py def _load_spec_dict(): spec_dict[key]['delta']=np.array(val("delta")) - np.array(static_root) static_root is the translation of SMPL root joint.

SMY19999 commented 2 months ago

Though I didn't do the experiment, I guess it is because the pose parameters are different between SMPL and SMPL-H. Just try using SMPL.

Hi,

Did you figure out why there is such displacement? I just have the same problem here. Besides, when I try to use the pose parameters in the "0000_body_info.json" located in the "poses" folder(e.g. tee_ZZHFVSA17Q_pants_straight_sides_ULX5THI3A5), it does not even return a valid SMPL-H body mesh...