Home

Awesome

<p align="center" width="100%"> <img src="assets/yagna.png" width="20%" alt="RITUAL"> </p>

RITUAL: Random Image Transformations as a Universal Anti-hallucination Lever in LVLMs

<!-- Arxiv Link, Project Link --> <div style='display:flex; gap: 0.25rem; '> <a href="https://arxiv.org/abs/2405.17821"><img src="https://img.shields.io/badge/arXiv-2405.17821-b31b1b.svg"></a> <a href="https://sangminwoo.github.io/RITUAL"><img src="https://img.shields.io/badge/Project%20Page-online-brightgreen"></a> <a href='LICENSE'><img src='https://img.shields.io/badge/License-MIT-blue.svg'></a> </div>

This repository contains the official pytorch implementation of the paper: "RITUAL: Random Image Transformations as a Universal Anti-hallucination Lever in LVLMs".

Updates

Overview

<p align="center" width="100%"> <img src="assets/overview.png" width="100%" alt="Overview"> </p> When conditioned on the original image, the probabilities for Blue (correct) and Red (hallucinated) responses are similar, which can lead to the hallucinated response being easily sampled. RITUAL leverages an additional probability distribution conditioned on the transformed image, where the likelihood of hallucination is significantly reduced. Consequently, the response is sampled from a linear combination of the two probability distributions, ensuring more accurate and reliable outputs.

Setup

conda create -n RITUAL python=3.10
conda activate RITUAL
git clone https://github.com/sangminwoo/RITUAL.git
cd RITUAL
pip install -r requirements.txt

Models

About model checkpoints preparation

Evaluation

About datasets preparation

Results

POPE

<p align="center" width="100%"> <img src="assets/pope.png" width="90%" alt="POPE results"> </p>

MME

MME-Fullset

<p align="center" width="100%"> <img src="assets/mme-fullset.png" width="90%" alt="MME-Fullset results"> </p>

MME-Hallucination

<p align="center" width="100%"> <img src="assets/mme-hallucination.png" width="90%" alt="MME-Hallucination results"> </p>

CHAIR

<p align="center" width="100%"> <img src="assets/chair.png" width="40%" alt="CHAIR results"> </p>

Examples

<p align="center" width="100%"> <img src="assets/llava_bench.png" width="100%" alt="LLaVA-Bench results"> </p> <p align="center" width="100%"> <img src="assets/llava_bench_appendix.png" width="100%" alt="LLaVA-Bench results"> </p>

Acknowledgments

This codebase borrows from most notably VCD, OPERA, and LLaVA. Many thanks to the authors for generously sharing their codes!

Citation

If you find this repository helpful for your project, please consider citing our work :)

@article{woo2024ritual,
  title={RITUAL: Random Image Transformations as a Universal Anti-hallucination Lever in LVLMs}, 
  author={Woo, Sangmin and Jang, Jaehyuk and Kim, Donguk and Choi, Yubin and Kim, Changick},
  journal={arXiv preprint arXiv:2405.17821},
  year={2024},
}