search

HugoTini / DeepBump

Normal & height maps generation from single pictures
https://hugotini.github.io/deepbump
GNU General Public License v3.0
1.07k stars 41 forks source link

Update for 3.1? #3

Closed melvixiodf closed 2 years ago

melvixiodf commented 2 years ago

Hi, Awesome work with this very powerful addon. Any chance for an update for 3.1?

Thanks for making this available for free to other artists.

Cheers,

HugoTini commented 2 years ago

Hello, thanks for the kind words, glad to hear the addon is useful to you ! As mentioned here, there is an issue with Blender 3.1 but it should be fixed automatically once onnxruntime is released for python 3.10.

melvixiodf commented 2 years ago

Ok. Thanks. Keep up the great work.

Btw, any chance you could add the ability to generate a shadow/ao removed albedo map to the addon? Is this possible with AI training? Love the results generated from the albedo map with shadows. Very realistic.

I also use the Generate height from Normal map addon as well.

HugoTini commented 2 years ago

Generating an AO map can be done by displacing a plane using the heightmap and baking the AO with Cycles.

Example An example of the original photo, DeepBump's normal, NormalHeight's height & Cycles' baked AO from height : ![maps](https://user-images.githubusercontent.com/65409496/176037770-983c0e2a-6220-4c9b-86c9-ff77ff124d97.png)

Now "delighting" the albedo from GI + hard & soft shadows is a whole research field with several approaches (Machine Learning, Differential rendering, etc.), that's for sure an interesting topic to work on if time would allow, but currently the addon has no such feature.

If you don't have hard shadows, one possible trick is to use the generated heightmap (and/or the AO map), invert it and blend it on the original picture in any image editor.

Example ![flat](https://user-images.githubusercontent.com/65409496/176037922-b844a6c6-a998-430c-adb0-b6ef3e640f91.png)

Back to the original issue, it should be solved with the next version of onnxruntime which is coming out mid-july

melvixiodf commented 2 years ago

Thanks for the examples. Very very Helpful.

Back to the original issue, it should be solved with the next version of onnxruntime which is coming out mid-july

That's great news. Thanks a ton.

Trass3r commented 2 years ago

It took some fiddling to install it manually:

3.2\python\bin\python.exe -m pip install flatbuffers sympy coloredlogs numpy==1.23.1
3.2\python\bin\python.exe -m pip install -i https://test.pypi.org/simple/ ort-nightly==1.12.0.dev20220711001
HugoTini commented 2 years ago

I have checked with Blender 3.2 today and it now installs without any issue