In order to make fine-tuning easier, I implemented a mechanism to easily save and load from edges bins that have already been computed.
What does it mean in practice:
You just have to run a command like the one in examples/tags_to_annotations/run_metadata_to_text_for_finetuning.sh to compute text bins from a new dataset that you want to fine-tune your model on.
What's here:
easily create a dataset from fine-tuning (guide on the README + code)
In order to make fine-tuning easier, I implemented a mechanism to easily save and load from edges bins that have already been computed. What does it mean in practice: You just have to run a command like the one in
examples/tags_to_annotations/run_metadata_to_text_for_finetuning.shto compute text bins from a new dataset that you want to fine-tune your model on.What's here:
Additionaly, I've improved
main.py:cc @sanchit-gandhi