Occlusion channel seems affect Diffuse

[junyiwuu]

Hi :) , In the paper, you mentioned that for k_orm you ignored/zero_out occlusion channel (ks first channel) in the code. after test the method on dozens of Megascan asset, I noticed that occlusion channel is actually calculating and (i guess ) somehow also affecting Diffuse optimization. It is not obvious on most asset, but do occasionally happen on some.

Please see example below: this asset here I only input roughness channel in k_orm, so the second green channel

The output:

and if i "add" the occlusion on DIffuse map, it looks like this: So feels like occlusion channel is taking away some black color that suppose on the diffuse map... but I thought we are not suppose to care about occlusion channel at all?

This is my json file: { "ref_mesh": "data/pile/pile.obj", "random_textures": true, "iter": 1000, "save_interval": 20, "texture_res": [ 1024, 1024 ], "train_res": [512, 512], "batch": 8, "learning_rate": [0.03, 0.003], "ks_min" : [0, 0.001, 0.0], "envmap": "data/irrmaps/aerodynamics_workshop_2k.hdr", "background": "white", "validate" : false, "learn_light" : false, "out_dir": "pile" }

Do you know why does this happen? Would really appreciate any suggestions. Thank you very much :)