Use code with OptiX
Hello, I have a project in my college that I have to code a denoising autoencoder and at the end I have to integrate it in the OptiX. Is there some way to do that?
Hello, I have a project in my college that I have to code a denoising autoencoder and at the end I have to integrate it in the OptiX. Is there some way to do that?

#1
Posted 12/12/2017 12:09 AM   
Depends in what you mean with "integrate in OptiX". OptiX is a high-level ray casting SDK. You define the behavior of different program domains like ray generation, closest hit, any hit, intersection, etc. inside your application using OptiX. Means you implement these programs normally using CUDA C++ and provide compiled PTX code of those to the OptiX API. If you just meant to feed synthesized images into your denoising autoencoder, that's the easiest thing. Just read the buffers from OptiX and do something on them. If that is implemented in CUDA you could even use CUDA-OptiX interoparability and save the copy operation. Note that the upcoming OptiX 5.0 release will contain a post-processing pipeline API which also features a built-in AI denoiser you might find interesting: [url]https://developer.nvidia.com/optix[/url]
Depends in what you mean with "integrate in OptiX".

OptiX is a high-level ray casting SDK. You define the behavior of different program domains like ray generation, closest hit, any hit, intersection, etc. inside your application using OptiX. Means you implement these programs normally using CUDA C++ and provide compiled PTX code of those to the OptiX API.

If you just meant to feed synthesized images into your denoising autoencoder, that's the easiest thing. Just read the buffers from OptiX and do something on them. If that is implemented in CUDA you could even use CUDA-OptiX interoparability and save the copy operation.

Note that the upcoming OptiX 5.0 release will contain a post-processing pipeline API which also features a built-in AI denoiser you might find interesting: https://developer.nvidia.com/optix

#2
Posted 12/12/2017 09:33 AM   
Scroll To Top

Add Reply