Not sure how bad it is these days, but a couple of years ago I was training a CNN and decided to just upload the data (images for training and testing) to my google drive folder - as I used google colab for development.
It was slow as molasses - unusable, really, and we're not talking about huge amounts of data either.
At least then, it seemed to be the google drive i/o that was the bottleneck, and the solution was to upload the training files to the colab session / VM.
It was slow as molasses - unusable, really, and we're not talking about huge amounts of data either.
At least then, it seemed to be the google drive i/o that was the bottleneck, and the solution was to upload the training files to the colab session / VM.