Open jamespoly862 opened 10 months ago
I don't know is it related but I notice there is lots of torch tensor is using cpu instead of gpu even it is okay to use, one example is here, it turns the image from ndarray to torch tensor in cpu, but it is absolutely fine to use cuda tensor.
I changed it to use cuda tensor instead and already reduce a certain amount of cpu cost and speed up the process by a lot.
Demo 3D inference give me an error on gpu_device = self.device on webcam_detector. Cancel this row, 3D inference function but give me high cpu consumption and low gpu consumption and too slowly. I have a nvidia 3070 ti and 2D inference work. I don't know why... CUDA is correctly installed.
Kind regards, James