What was the training / eval dataset?
#2
by wendlerc - opened
Was the training/eval dataset something like this: https://huggingface.co/datasets/AdamCodd/Civitai-8m-prompts ?
No, this one was trained on a (small) personal dataset collected on internet and doesn't include any generated images.
@AdamCodd Thanks for your work and sharing. Can you please share what's the cutoff boundary that is used for the classification here ?
Hello @abhishek072 ,
The cutoff boundary used for classification is 0.5. Any image with a predicted NSFW probability greater than 0.5 is classified as NSFW, while others are classified as SFW.