nazounoryuu/vlm-htr · Apply for community grant: Academic project (gpu)

This is a project tied to my master' thesis.
In this project, I fine-tune Florence-2, a visual language model, on several tasks: text region detection, line detection, line segmentation, and text recognition. Then, I compare a pipeline made of two or three fine-tuned Florence model with a traditional HTR pipeline on the text recognition task for historical Swedish manuscripts.

Since it's a VLM, Florence-2 is very slow on CPU, so it would be great if I can have access to GPU to speed up inference.