search

pixelite1201 / BEDLAM

226 stars 18 forks source link

Some unclear information of BEDLAM SMPLX annotations #55

Closed mks0601 closed 2 months ago

mks0601 commented 2 months ago

Hi,

Thanks for your amazing work. Could you help me to clarify some information about your dataset?

  1. The facial expression code of SMPL-X fittings are not provided. The paper says they are set to zero, so can I simply set them to zero when using this dataset?

  2. I'm using SMPL-X fitting in Images and GT labels of Synthetic Image Data at 6fps used for training and testing in the Download section of the BEDLAM webpage. It seems they provide gender information, but if I use the gendered smplx layer, the projection is wrong. Instead, if I use neutral smplx layer by ignoring the provided gender information, the projections are correct. So should I ignore the gender information of those annotation files and use neutral gender?

  3. It seems the file of the second question incldues AGORA images and annotations. Do all AGORA images are included?

pixelite1201 commented 2 months ago

Hello,

  1. Yes, you could set facial expression to zero.
  2. The provided fitting in 'Images and GT labels at 6fps' is what we used in trainingBEDLAM-CLIFF and consist of parameters for neutral body. The parameters are obtained by fitting SMPL-X neutral body model to gendered body with minimal error. If you want to use gendered bodies instead of neutral, we also provide SMPL-X gendered bodies in Download section.
  3. We didn't provide crops of kids for AGORA annotations. It consists of only adult bodies. If you are using AGORA images, please agree to the AGORA license as well.