EPFL-VILAB / omnidata

A Scalable Pipeline for Making Steerable Multi-Task Mid-Level Vision Datasets from 3D Scans [ICCV 2021]
Other
395 stars 49 forks source link

What if I put the 512*512 image to this model... #17

Closed lxl1115 closed 1 year ago

lxl1115 commented 1 year ago

Thank you for this wonderful work and it really helps me alot! I have a question here: my images are 512 resolution and I want the 512 resulution normal map. I think if I just simply interpolate the output normal map from 384 resolution to 512 resolution will cause a huge error since the normal is very sensitive. So i'm wondering if I can just put the 512 images into the model, will I get a wrong normal map? I compared them by visualization and I cant figure out...

alexsax commented 1 year ago

You can use 512x512 images, or you can resize using nearest neighbor interpolation: both should work!