https://github.com/aashish75/blip-pathvqa-peft

https://github.com/aashish75/blip-pathvqa-peft

Science Score: 13.0%

This score indicates how likely this project is to be science-related based on various indicators:

  • CITATION.cff file
  • codemeta.json file
    Found codemeta.json file
  • .zenodo.json file
  • DOI references
  • Academic publication links
  • Academic email domains
  • Institutional organization owner
  • JOSS paper metadata
  • Scientific vocabulary similarity
    Low similarity (4.8%) to scientific vocabulary
Last synced: 10 months ago · JSON representation

Repository

Basic Info
  • Host: GitHub
  • Owner: Aashish75
  • Language: Jupyter Notebook
  • Default Branch: main
  • Size: 521 KB
Statistics
  • Stars: 1
  • Watchers: 1
  • Forks: 0
  • Open Issues: 0
  • Releases: 0
Created over 1 year ago · Last pushed about 1 year ago
Metadata Files
Readme

README.md

blip-PathVQA-PEFT: Fine-Tuning BLIP on PathVQA with LoRA

This repository contains code for fine-tuning the BLIP model on the PathVQA dataset using Parameter-Efficient Fine-Tuning (PEFT) with LoRA (Low-Rank Adaptation). The goal is to optimize BLIP for pathology-based Visual Question Answering (VQA) while reducing computational overhead.

🚀 Project Overview

  • Fine-tunes Salesforce/blip-vqa-base on the PathVQA dataset.
  • Implements LoRA (Low-Rank Adaptation) instead of full fine-tuning to reduce memory usage while maintaining strong performance.
  • Compares LoRA fine-tuning vs. traditional full fine-tuning.
  • Evaluates model performance before and after fine-tuning.

📂 Dataset: PathVQA

  • The dataset is loaded from the Hugging Face datasets library: ```python from datasets import loaddataset dataset = loaddataset("flaviagiammarino/path-vqa")

📊 Evaluation Results

The training significantly increased the capabilities of the base BLIP model in pathological visual question answering (yes/no questions).

🔹 After LoRA Fine-Tuning

| Metric | Score | |-------------|----------| | Accuracy | 85.19% | | Precision | 85.89% | | Recall | 86.84% | | F1 Score | 86.36% |

🔹 Before Fine-Tuning (BLIP Base Model)

| Metric | Score | |-------------|----------| | Accuracy | 51.64% | | Precision | 53.44% | | Recall | 81.22% | | F1 Score | 64.47% |

🔹 Overall Improvements

| Metric | Increase | |-------------|-------------| | Accuracy | +33.55% | | Precision | +32.45% | | Recall | +5.62% | | F1 Score | +21.89% |

LoRA significantly improves performance while being computationally efficient!

Owner

  • Name: Aashish M
  • Login: Aashish75
  • Kind: user

I code

GitHub Events

Total
  • Watch event: 1
  • Push event: 16
  • Create event: 2
Last Year
  • Watch event: 1
  • Push event: 16
  • Create event: 2