Request for Pre-processed Data Sample for Virtual Try-on Demo

[Vikas-U-Badiger]

I'm a student who has successfully built the Docker environment for Gaussian Wardrobe. I'm trying to run the virtual try-on demo (--mode=exchange_cloth), but I'm stuck at the data preprocessing stage.

I have the raw 4D-DRESS dataset, but the preprocessing pipeline requires many intermediate files that I don't have. Specifically, the scripts gen_weight_volume.py and gen_pos_maps.py are failing because they can't find files like smpl_params.npz.

I would like to request if it's possible to share a small, pre-processed test sample (e.g., one subject, one garment layer, a few frames) suitable for the exchange_cloth script. Having such a sample would allow me to validate the pipeline for my academic project.

Specifically, I'm looking for:

smpl_params.npz

cameras.pkl

smpl_pos_map_offset_body/ and smpl_pos_map_offset_cloth/

(Optional) A pre-trained checkpoint

Thank you for your fantastic work and for your consideration.

[azuxmioy]

Hi Vikas,

Thanks for reaching out. We'll take a look and keep you posted here

@zhiyi-chen217

[zhiyi-chen217]

Hi Vikas,

Thanks for the comment!

• I have added the checkpoints in the README.
• Here is a small template you can start from for the exchange cloth data.
  • Most of the files needed are already there. You just need to generate the pose maps
  • Here I also included a sequence of poses you can try "00127_train.npz"
  • Here, I included a 134_140.yaml file you can use as the config, you can tweak with the smpl_pos_map, sequence_name, exchange_cloth->change_mode and checkpoints to experiment with different configurations