I was inspired by graphical representation of certain data like sound and binary files to use Fastai vision for classification, so I tried IMDB dataset: converted the first 784 characters to numbers and converted to image (224x224) with each char repeated 64 times. I know this is not the best setup or approach, but do you think this is supposed to work in similar situations (ex. topics, authors, …). My result after a few epochs (using resnet34 - I know not the best) was a little better than random!
Anybody tried something similar?
