I have an RTX 2070 and a Ryzen 2700X on windows 10 and am trying to compile and run the NGX SlowMo sample.
It compiles successfully, I have the latest NVIDIA drivers, I installed the NGX SDK, and I installed the CUDA toolkit.
However, when I actually run the SlowMo sample it initialises and finishes very quickly, producing an output file that is 0 bytes. When running from the VS 2017 debugger it catches an access violation trying to read a null pointer in the function NVSDK_NGX_CUDA_EvaluateFeature.
The only null pointer I have been able to locate in the sample is the scratch buffer, due to NVSDK_NGX_CUDA_GetScratchBufferSize returning a 0 scratch buffer size and the sample skipping the allocation of the scratch buffer in such a case. I have not made any changes to the sample, and all functions are giving successful return codes.
I have tried with multiple video files, firstly an flv at 1080p60 that is 4m49s long, then an mp4 at 1080p60 (transcoded from the flv), and finally another mp4 at 1358x1000 at 60fps that was explicitly encoded using FFmpeg.
When I comment out the call to NVSDK_NGX_CUDA_EvaluateFeature the sample successfully creates a video, it just contains many black frames since the AI was never actually run.
The InPainting sample seems to work correctly.
Is there something wrong with my setup? Or is there a problem in the sample or the SDK itself for SlowMo?