magvit2-pytorch
magvit2-pytorch copied to clipboard
lucidrains
Pixelated image reconstruction
Trained the model on a set of 10K images (randomly cropped 128 x 128 patches) for 6K iteration with a batch size of 8. The output looks pixelated. Is this what is expected, or the output should somehow look similar to the input? Any feedbacks are appreciated!
@coolbunnyx is there any chance you could share the script you used to train this?
meeting the same problem，Have you solved this problem？ @coolbunnyx
I've given it up; not sure if the issue is the code or just me lacking the computing resources to make it converge quickly to the right solution. Intuitively, it should converge to something similar to the input image instead of getting pixelated, though I might be wrong.
Trained the model on a set of 10K images (randomly cropped 128 x 128 patches) for 6K iteration with a batch size of 8. The output looks pixelated. Is this what is expected, or the output should somehow look similar to the input? Any feedbacks are appreciated!
Could you please share your config for training? I just cannot reproduce anything except the black blocks. Many thanks!
Trained the model on a set of 10K images (randomly cropped 128 x 128 patches) for 6K iteration with a batch size of 8. The output looks pixelated. Is this what is expected, or the output should somehow look similar to the input? Any feedbacks are appreciated!
Could you please share your config for training? I just cannot reproduce anything except the black blocks. Many thanks!
It may work if you turn off the use_gan.
@lucidrains @Borhanv-nd I'm running into the same situation, do you have any comments on this issue? Do I need to train more steps?
