Closed gngdb closed 9 years ago
Made this notebook, calle dit Log loss over image size
. There is a small correlation between image size and log loss, but probably nothing significant. Looked at a rolling average and linear regression. Also looked at the results of the parallel model to see if it might have performed better, but results appear to be exactly the same.
Actually, I should really have looked at models that were trained on 64 by 64 images to see if they're even better. If I have time I'll come back to this.
Do the big images have big log loss. Are we losing a lot by resizing or not?