Master Yoda tell you what's in the image!
Try it at
https://yodacaptioner.up.railway.app/
More details about the model can be found at:
https://huggingface.co/vkao8264/blip-yoda-captioning
This repo contains the scripts to generate the captions used to fine-tune the BLIP image-to-text model, and the training script.
For the web app itself, the repo is here
In order to run the scripts, you will need to have the COCO captions dataset on your machine
captions_generator.py generates 30000 Yoda-style captions using phi3 LLM imported from lmphi.py
blip_training.py then pairs the captions with images and fine-tunes the BLIP model