Linaqruf / kohya-trainer

Adapted from https://note.com/kohya_ss/n/nbf7ce8d80f29 for easier cloning
Apache License 2.0
1.85k stars 307 forks source link

Caption using Coca instead of BLIP #74

Open DanielShemesh opened 1 year ago

DanielShemesh commented 1 year ago

Now that we have open-sourced COCA which is SOTA at image captioning, I think it's better to use it instead of BLIP. https://huggingface.co/spaces/laion/CoCa

artificialguybr commented 1 year ago

Git still better for some things. Git and Coca are almost the same now.

Linaqruf commented 1 year ago

Thank you! i'll consider to add that in the future updates, but for now i'll revert Git to blip because of slower inference