-
|
Hello! I ran the code in pix2pix.ipynb for the facades data and noticed that my results were much worse than results from the pretrained model's. Has anybody been able to replicate the image quality of the pretrained models, and if so, how did they do it? (Ideally the authors will chime in, but I'd love to hear from anybody!). Thank you! |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 2 replies
-
|
Found the answer – I wasn't using the author's suggested params after all! Instead of their batch size = 1, I was using batch size = 16 and getting worse results. Using all of their code out of the box without modifications worked. |
Beta Was this translation helpful? Give feedback.
-
|
Hello! So what parameters did you finally use to achieve good results with batch size 16 ? |
Beta Was this translation helpful? Give feedback.
-
|
I was not able to make it work with bs=16. I ended up using bs=1 and all of the other parameter values suggested by the authors. |
Beta Was this translation helpful? Give feedback.
Found the answer – I wasn't using the author's suggested params after all! Instead of their batch size = 1, I was using batch size = 16 and getting worse results. Using all of their code out of the box without modifications worked.