search

SamsungLabs / DINAR

Inference code for "DINAR: Diffusion Inpainting of Neural Textures for One-Shot Human Avatars"
Other
100 stars 6 forks source link

About prepared data #6

Open breeze0011 opened 11 months ago

breeze0011 commented 11 months ago

Hello, I tried to put the prepared data's PKL files into Blender to display the mesh, but I didn't succeed. If possible, could you upload the prepared data's mesh results generated by SMPL-X? I wonder whether the results generated by SMPL-X myself are correct.

david-svitov commented 11 months ago

Hello! Could you please explain how exactly you are trying to open PKL files in Blender? PKL files do not contain meshes, but contain SMPL-X parameters. You can prepare your PKLs in a similar way to mine, for example by viewing their contents in Python using the Pickle library. If you have questions about the compatibility of SMPL-X parameters, there is a discussion here: https://github.com/SamsungLabs/DINAR/issues/5

breeze0011 commented 11 months ago

Thank you for your reply. There were some issues with the results generated by SMPL-X myself before, and I have found the error.

mayank64ce commented 10 months ago

Hey @breeze0011 what kind of errors were you facing ? I am getting these from smplifyx : https://github.com/vchoutas/smplify-x/issues/210

breeze0011 commented 10 months ago

image I met this problem too. When you use smplx, you need to pay attention to the folder. You can try put only one image in this folder. image

mayank64ce commented 10 months ago

Yeah. What's happening is if I put any image from any other dataset along with snapshot people, it somehow morphs the snapshot outputs as well. So you are saying if I use only images from same dataset, I shouldn't have any problems ?

breeze0011 commented 10 months ago

Sorry, I don't know why this situation occurs. I think the images in this folder will affect each other during the generation process.