vq-vae-2-pytorch icon indicating copy to clipboard operation
vq-vae-2-pytorch copied to clipboard

Published 20 hours ago • verified publisher icon rosinality

Struggling to figure out why model learns for 3 images but not 4

Open steve-1820 opened this issue 5 years ago • 1 comments
trafficstars

Hey guys,

This might be a silly question but I was testing this out on images of sneakers. Works well for 1, 2 or 3 sneakers but for some reason it never learns anything for 4 images or above.

image

I've tried multiple variations of images to no success.

Any ideas?

steve-1820 avatar Feb 09 '20 11:02 steve-1820

Hi Steve,

Are those images created with the python train_vqvae.py [DATASET PATH] command? If so I think that you have specified for the evaluator to only reconstruct 3/4 images to view. The images on the left are the input images and the images on the right are the reconstructed images. If you train for a few more epochs you should have much better results.

harrygcoppock avatar May 25 '20 16:05 harrygcoppock