I am relatively very new to DL. I am thinking of implementing a code like below
… predict using a pretrained model and output is an image. These images are fetched as frames from videos
… use the predicted output to train another model
I have only one GPu 8gb. I am wondering is there a way to do data fetching and training together on a GPU without bottleneck. Is there a way to optimitze them.
However I tried to save individual frames and also tensor.pt file. Both seems to take more time for preprocessing and training. I am using pytorch and any suggestion would be greatly helpful. Thanks