I'm optimizing by initializing with existing geometry with UV map and the base geometry being used is a human facial model. I am observing that while the materials appear right at the end of optimization, some facial features are lost at this point. Specifically eye and lip details. Below the paramters used for optimization:
{ "base_mesh": "data/face/mesh.obj", "ref_mesh": "data/face/mesh.obj", "random_textures": true, "iter": 5000, "save_interval": 100, "texture_res": [ 1024, 1024 ], "train_res": [1024, 1024], "batch": 4, "learning_rate": [0.03, 0.01], "ks_min" : [0, 0.1, 0.0], "ks_max" : [0, 1.0, 1.0], "envlight": "data/irrmaps/aerodynamics_workshop_2k.hdr", "validate" : false, "lock_pos" : true, "display": [{"latlong" : true}], "background" : "white", "denoiser": "none", "no_perturbed_nrm" : true, "n_samples" : 8, "out_dir": "face" }
It would appear that these missing features are being treated as noise ? Are there some assumptions in the material estimation stage that would lead to such behaviour ?
I'm optimizing by initializing with existing geometry with UV map and the base geometry being used is a human facial model. I am observing that while the materials appear right at the end of optimization, some facial features are lost at this point. Specifically eye and lip details. Below the paramters used for optimization:
{ "base_mesh": "data/face/mesh.obj", "ref_mesh": "data/face/mesh.obj", "random_textures": true, "iter": 5000, "save_interval": 100, "texture_res": [ 1024, 1024 ], "train_res": [1024, 1024], "batch": 4, "learning_rate": [0.03, 0.01], "ks_min" : [0, 0.1, 0.0], "ks_max" : [0, 1.0, 1.0], "envlight": "data/irrmaps/aerodynamics_workshop_2k.hdr", "validate" : false, "lock_pos" : true, "display": [{"latlong" : true}], "background" : "white", "denoiser": "none", "no_perturbed_nrm" : true, "n_samples" : 8, "out_dir": "face" }
It would appear that these missing features are being treated as noise ? Are there some assumptions in the material estimation stage that would lead to such behaviour ?