ashual
ashual
unfortunately I don't have the original checkpoints anymore. But I think continue training will just cause overfitting.
1. only test set 2. no, only once. The results indeed are slightly different in each run (might be slightly better or slightly worst)
@viperit thanks for notifying me I've made a massive refactor to the code before publishing and while it wasn't supposed to change the results, it may have I'll train it...
We cite [28] a lot in the experiments section. Is there a specific reference that you think is wrong?
yes, you are absolutely right. I'll fix it in the Arxiv version, Thanks for letting me know
does anyone know why they don't use .eval()? is it a bug or is it by design?
it might be because the exclamation mark. try it this way: `wget -O outputs/checkpoints.zip "https://1drv.ws/u/s!AjjUqiJZsj8whLNyoEh67Uu0LlxquA?e=dkOnhQ"`