WeiPhil / nbvh

Source Code for "N-BVH: Neural ray queries with bounding volume hierarchies" (SIGGRAPH 2024)
MIT License
109 stars 7 forks source link

The screen shows black without chess scene? #3

Open leftstone2015 opened 2 days ago

leftstone2015 commented 2 days ago

hi, thanks for your great work.

I am trying to run the code in my PC with the chess scene.
Following the instructions in readme, it seems that it goes well from the log. The log as follows:

[Info] Loaded Envmap from ../nbvh/scene_data\hdris\photo_studio_loft_hall_4k.exr [Info] Loading neural BVH from : ..//nbvh/scene_data/scenes_path_tracing/chess\nbvhs\nodes_32940_hash_18_lvl_8_tsteps_10000.nbvh... [Info] Loaded neural LibBVH with 25249 nodes. [Info] Deserializing network model from : ../nbvh/scene_data/scenes_path_tracing/chess\network_models\nodes_32940_hash_18_lvl_8_tsteps_10000_network.json [Info] Done! [Info] Done! [Info] Done! [Info] Added renderer : Neural BVH Renderer

But the screen shows black without chess scene, as follows: image

Could you help check it and tell me whether it it OK? What is next step to show the scene?

WeiPhil commented 2 days ago

Hi, This looks fine. Have you clicked on "Run inference" in the UI? Best, Philippe

leftstone2015 commented 2 days ago

Thank you for replying quickly. I've tried it, but the screen still goes back. By the way, the frame rate dropped for 1000+fps to 10-fps.

WeiPhil commented 2 days ago

Something is probably going wrong with the display then. Are you using the binary release or did you compile from source? If you are using the binary release I would start by building from source. I would also need info on your configuration to reproduce this (GPU + driver, Cuda version, operating system, C/C++ compiler)

leftstone2015 commented 1 day ago

Here are the information of my PC:

  1. GPU: NVIDIA GeForce RTX 3090
  2. Driver:NVIDIA-SMI 560.94 Driver Version: 560.94 CUDA Version: 12.6
  3. OS: Win11
  4. cmake cmd: cmake -G "Visual Studio 17 2022" -DCMAKE_CUDA_ARCHITECTURES=86 -A x64 -B build

cmake output: Selecting Windows SDK version 10.0.22621.0 to target Windows 10.0.22631. -- The C compiler identification is MSVC 19.38.33141.0 -- The CXX compiler identification is MSVC 19.38.33141.0 -- The CUDA compiler identification is NVIDIA 12.6.68

By the way, if I use -DCMAKE_CUDA_ARCHITECTURES=89, I will get the link error of "CUDA_SYNC_CHECK".

WeiPhil commented 1 day ago

This is likely a mismatch between compute capabilities. Make sure to clean the build directory completely and start te compilation from scratch specifying -DCMAKE_CUDA_ARCHITECTURES=86 (a clean from scratch config and rebuild is necessary as some files might still be using 89 since you specified this). Note that your 3090 expects compute capability 86 as specified here. If that still doesn't solve the issue I would try an older version of CUDA, e.g. 12.0 for which our build was tested.

leftstone2015 commented 1 day ago

When I use the binary, it goes well.

Something is probably going wrong with the display then. Are you using the binary release or did you compile from source? If you are using the binary release I would start by building from source. I would also need info on your configuration to reproduce this (GPU + driver, Cuda version, operating system, C/C++ compiler)

leftstone2015 commented 1 day ago

This is likely a mismatch between compute capabilities. Make sure to clean the build directory completely and start te compilation from scratch specifying -DCMAKE_CUDA_ARCHITECTURES=86 (a clean from scratch config and rebuild is necessary as some files might still be using 89 since you specified this). Note that your 3090 expects compute capability 86 as specified here. If that still doesn't solve the issue I would try an older version of CUDA, e.g. 12.0 for which our build was tested.

I use compute capability 86 from scratch, it does not work. The screen is black.