Open abdullahkhan-z opened 4 years ago
Hi @abdullahkhan-z ,
We used a dataset (human seg) which has per-face segmentation labels at a high resolution. To make learning easier, we simplified the meshes to a lower resolution. Then we needed to run an algorithm to calculate the new segmentation labels on the low resolution meshes (which came from the high resolution version with labels).
If your segmentation class is already relatively low resolution, you can skip the simplification stage.
I want to simplified my own meshes to a lower resolution too. Could you help me about the algorithm to calculate the new segmentation labels on the low resolution meshes please?
This is what is stated in wiki.
What I understood is that, for original mesh, annotate edges corresponding to their class as per format of .eseg file, that I might be able to do somehow or perhaps you have some scripts to do so. According to above quoted statement, the mesh number of edges has to be consistent with the rest of meshes that means there would be quality downgrading or upgrading something like that and from what you stated above, the annotated file will be re evaluated by some script to handle that down sampled or up sampled mesh. I saw the .eseg files in the dataset, from the looks of. it, I was wondering, why can't I simply create all meshes of same number of edges and create a .eseg file based on the edges I want to be part of a particular class?