While training the GPU utilization is nearly 94% even with a batch size of 4. Even after reducing the size of images by 50%, the GPU utilization isn’t decreasing. Is there anything wrong with the code?
model = fasterrcnn_resnet50_fpn()
I’m not sure I understand the issue correctly, but why do you want to decrease the GPU utilization? If you want to slow down GPU util. and thus slow down your code (maybe for debugging purposes?) you could use time.sleep calls inside your training code.
sorry, maybe my wording is wrong. When I use batch size 4 with normal images and batch size 4 with images of reduced size, the memory utilized by GPU is the same. I want to increase my batch size.