- Support for more caption models. In addition to blip-base and blip-large there is now blip2-2.7b (15.5GB), blip2-flan-t5-xl (15.77GB), git-large-coco (1.58GB). See run_gradio.py for example.
- Newly exposed class LabelTable and functions list_caption_models, list_clip_models, load_list
- See README for example how to rank your own list of terms using LabelTable