The following notebooks are designed for local environments, as processing and analyzing datasets on cloud may not be efficient.


- preprocess_bcdrf.ipynb: notebook to process the F versions of BCDR dataset to the .tsv format used by BiomedGPT.

	1. Download and unzip the BCDR datasets, select and process any F version. 
	2. Create your own questions (open and closed) regarding the lesion descriptors. 
	3. Run the notebook to generate and verify the script.

- preprocess_omnivqa.ipynb: notebook to process the OmniMedVQA datasets to the .tsv format used by BiomedGPT.

	1. Download and unzip OmniMedVQA (https://huggingface.co/datasets/foreverbeliever/OmniMedVQA/tree/main).
	2. Install necessary libraries. 
	3. Run the script, selecting the dataset you want to process. 
	4. This will create a datasets folder containing the folders of processed datasets. 

- vqa_zeroshot_analysis.ipynb: notebook that analyzes the results from running gcloud_zeroshot_vqa.ipynb in Colab. You can also analyze results from gcloud_finetune_evaluate_vqa.ipynb.

	1. Run zeroshot on BiomedGPT using the zeroshot script on Google Colab with the test.tsv file.
	2. Save the test.tsv file, the logs and json files on a 'zeroshot_results' folder within the 'datasets'. 
	3. Run the notebook. This will generate an analysis folder within 'zeroshot_results' with the images.

- vqa_finetune_analysis.ipynb : simple notebook to process the log file resulting from the gcloud_finetune_vqa.ipynb notebook.   

	1. Finetune a dataset using gcloud_finetune_vqa.ipynb.
	2. Save the .log file from finetuning on a 'finetune_results' folder within the 'datasets' folder.
        3. Evaluate the model using gcloud_finetune_evaluate_vqa.ipynb. Save the logs, json and test.tsv files. 
	3. Run the notebook. This will generate graphs displaying results. 


 