Use Yolo5 to run your own data set, torch_GPU and CUDA are installed, use GPU to run the model, the GPU occupancy rate is 1%, set epoch to 300, batch size to 32, about 6min more/epoch, the data set is very small , Just over a thousand. I checked related articles on the Internet and found that it may be caused by reading data. Change the action='store_true' of cache-image to 'store_false'.
changed to
After modification, each epoch is about 13s! Super fast, the GPU occupancy rate becomes 33%, and the CPU occupancy rate is 50%.
I will find a way to change it later, it is very difficult for a computer novice to learn, and the thesis is very difficult (wailing~)!