Closed constantm closed 1 year ago
I got it working with the Blender script located in the Nvdiffrec Monte Carlo repo here: https://github.com/NVlabs/nvdiffrecmc/blob/main/blender/blender.py
I've been generating synthetic datasets with Blender, using COLMAP to estimate camera poses from around 100 images, and then running it through nvdiffrec to get back a mesh + textures. The mesh output is pretty decent, however I'm having difficulty getting good textures. The textures look very washed out. If I look at the images being saved during training, the textures look pretty good. However, when I open the finished me
hi, could you please give me some details about how to use the script 'blender.py' in Blender? I haven't use this software before, and I got stuck here, thanks!
@sadexcavator it's pretty straight forward:
blender.py
script in BlenderThe result should be a mesh in your workspace with shading nodes set up correctly.
Hello! Firstly, this is awesome work and I've loved playing around with it so far.
I've been generating synthetic datasets with Blender, using COLMAP to estimate camera poses from around 100 images, and then running it through nvdiffrec to get back a mesh + textures. The mesh output is pretty decent, however I'm having difficulty getting good textures. The textures look very washed out. If I look at the images being saved during training, the textures look pretty good. However, when I open the finished mesh, the textures look very washed out. Below I've attached a few crops to detail my issue:
Crop of input image in
img_mesh_pass_000100.png
:Crop of training preview in
img_mesh_pass_000100.png
- what I would expect the end result to look like:Crop of mesh imported to Blender with normal and specular maps applied:
The Blender node setup used:
Opening the generated mesh in Meshlab has the same washed-out look:
I would expect the final output to look like the training previews (image 2), but I might be misunderstanding what the training previews actually are.
My config is as follows:
So, my questions are:
Any input here would be greatly appreciated, thank you!