camaro
camaro
Wow, was it not codec problem?? Thanks, I'll try it!
+1 here!
I believe you can use it by just replacing BCEWithLogitsLoss in your code with this variant. https://github.com/bamps53/SeesawLoss/blob/476d60f6c8528f26491a34e397396b2b06ed380d/seesaw_loss.py#L48 But please note that it's now obsolete as original authors updated from 1st...
Yes, this one is the original implementation actally. https://github.com/bamps53/SeesawLoss/blob/476d60f6c8528f26491a34e397396b2b06ed380d/seesaw_loss.py#L48 Sorry for the confusing class name, I was miunderstood at first so I named SeesawLoss to the one which requires class...
> so you don't need else No, this condition is true only for the very first batch, from 2nd batch class_count is accumulated one. > what about compensation factor in...
As same as original repo, I use ImageNet mean and std normalization.
Thanks, that's my mistake! I made my code be able to run on kaggle , colab or my local GPU. My code assume directory like that: root |- input/ |-...
Hi @sayakpaul, I checked #6729 and attempted to run your test notebook (https://huggingface.co/sayakpaul/test-model-card-template-dreambooth/blob/main/test_dreambooth_model_card.ipynb). However, it appears to not work for #6729 as it uses the `instance_prompt` keyword arguments in `load_or_create_model_card`....
Oh, it seems not only notebooks, but also training scripts in examples are not workings now. I made PR #6895 to fix these before this issue.
@sayakpaul Thanks, I got it! I'll split the PR for each.