Hello! I'm trying to run a counter with CUDA to take the load off the CPU
Installed all the necessary dependencies (CUDA, CuDNN, torch, torchvision, torchaudio), torch.cuda.is_available() returns True, but when people-counter.py is run, the process is not displayed in nvidia-smi.
In gpustat I see that all the load goes to the CPU.
Has anyone managed to do this?
Hello! I'm trying to run a counter with CUDA to take the load off the CPU Installed all the necessary dependencies (CUDA, CuDNN, torch, torchvision, torchaudio), torch.cuda.is_available() returns True, but when people-counter.py is run, the process is not displayed in nvidia-smi. In gpustat I see that all the load goes to the CPU. Has anyone managed to do this?