NVlabs / nvdiffrec

Official code for the CVPR 2022 (oral) paper "Extracting Triangular 3D Models, Materials, and Lighting From Images".
Other
2.09k stars 222 forks source link

some suggestions why my "lego" mesh and textures looks so terrible in blender? #127

Open miseon119 opened 1 year ago

miseon119 commented 1 year ago

I used nerf lego synthetic dataset. my nerf_lego.json is

{
    "ref_mesh": "data/nerf_synthetic/lego",
    "random_textures": true,
    "iter": 5000,
    "save_interval": 100,
    "texture_res": [ 2048, 2048 ],
    "train_res": [800, 800],
    "batch": 8,
    "learning_rate": [0.03, 0.01],
    "ks_min" : [0, 0.25, 0.0],
    "dmtet_grid" : 128,
    "mesh_scale" : 2.4,
    "laplace_scale" : 3000,
    "display": [{"latlong" : true}, {"bsdf" : "kd"}, {"bsdf" : "ks"}, {"bsdf" : "normal"}],
    "background" : "white",
    "out_dir": "nerf_lego"
}

my mesh result is:

image image

What's wrong with my parameters? Is it that I used too small "dmtet_grid" value?

And "ethiopianHead_rescaled" dataset's result is also bad.

image image image
mints7 commented 1 year ago

I'm very sorry that I cannot solve your problem. And I need you some help. How did you obtain "nerf_lego.json"?

mspark2023 commented 1 year ago

I'm very sorry that I cannot solve your problem. And I need you some help. How did you obtain "nerf_lego.json"? I'm not sure about your question, maybe you mean this file?

huohuohuohuohuohuohuohuo commented 1 year ago

Thanks for your amazing work ! But I got the same problem. The dataset and .json are exactly set following the project, however the generated meshes are terrible. Btw, I used dr.RasterizeCudaContext(). I don't know if this change will affact the quality of mesh. 2023-06-16 09-24-16 的屏幕截图

2023-06-13 08-41-33 的屏幕截图 I just drag the .obj into the meshlab.