The repository is written for AutoGenDA.

extract_data.sh extracts the data according to the imbalanced or low-shot setting.
textual_inversion.sh fine-tune a diffusion model on the target data.
merge_emb.sh merge the learned embedding.
generate_caption.sh generate image captions.
gen_autodiff_aug.sh generate AutoGenDA augmented images.
train_clf.sh traina classifier on the augmented iamges.

Referenece
The fine-tuning code and the semantic_aug module are adapted from the DAFusion work.
