CUFFS dataset release

[SWWdz]

Thanks for your work! I have a problem annotating the SMPL-X parameters. Can you tell me how to annotate my image with SMPL-X parameters.

[fabienbaradel]

Hi @SWWdz . It seems to be not very related to this repo, please have a look to the SMPLX repo you should find more information about how to annotate your image with SMPL-X parameters :)

[SWWdz]

You mentioned the Synthetic whole-body CUFFS dataset in your article. How can I download the data set?

[fabienbaradel]

Oh I see, sorry. For the moment I cannot give you a date for the release. I will keep you updated as soon as we release the data.

[fabienbaradel]

Hi @SWWdz , CUFFS dataset is now available here.
The zip file contains images and a pkl file with annotations. I will provide soon more info about how to process it. Stay tuned.

[SWWdz]

Thanks for your share!

[nehc0]

Hi @SWWdz , CUFFS dataset is now available here. The zip file contains images and a pkl file with annotations. I will provide soon more info about how to process it. Stay tuned.

Hi @fabienbaradel , the only pkl file contained in the zip file is humgen3d_EFH_like_all.pkl , which is empty. Is the annotations not available yet?

[fabienbaradel]

The new zip file now contains images, license and annotation file called cuffs_annots_20240905.pkl.
You can find SMPL-X parameters for each human in this directory.
I will release soon the associated dataloader, stay tuned.

[nehc0]

The new zip file now contains images, license and annotation file called cuffs_annots_20240905.pkl. You can find SMPL-X parameters for each human in this directory. I will release soon the associated dataloader, stay tuned.

That's great, thanks! I still have a question: is it that the dataset doesn't have facial expression parameters as in BEDLAM? I found the parameters in that pkl file don't consist of expression parameters.

[fabienbaradel]

Yes you are right.

[SWWdz]

Do you use the optimization method to get smplx parameters from your 2D observation? I am interested in how to get the GT label parameters of humans in smplx parametric model.

[fabienbaradel]

@SWWdz , the pipeline for generating imgaes are annotations is described in the paper (see supp. mat.), I encourage you to look at the CUFFS dataset section to get all the details. In short, we sample SMPL-X parameters, generate the SMPL-X mesh and do a retargetting to Human Generator mesh using optimization and finally do the rendering. I hope that my answer helps for your work.