ONLY

KO | EN

gitlite — search

#javascript #python #hacktoberfest #react #ai #typescript #llm #go #golang #android #machine-learning #rust #deep-learning #linux

(★ 51)

[ICCV 2025] ONLY: One-Layer Intervention Sufficiently Mitigates Hallucinations in Large Vision-Language Models

ONLY 최신버젼 다운로드

최종 버전 다운로드 (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# [ICCV 2025] ONLY

[![Website](https://img.shields.io/badge/Project-Website-green)](https://zifuwan.github.io/ONLY/) [![arXiv](https://img.shields.io/badge/arXiv-2507.00898-red)](http://arxiv.org/abs/2507.00898) [![Conference](https://img.shields.io/badge/ICCV-2025-blue)](https://iccv.thecvf.com/) [![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)

## 👀Introduction

This repository contains the code for our ICCV 2025 paper `ONLY: One-Layer Intervention Sufficiently Mitigates Hallucinations in Large Vision-Language Models`.

## 💡Environment

We test our codebase with PyTorch 2.0.1. Please install the corresponding PyTorch and CUDA versions according to your computational resources.

```
conda create -n ONLY python=3.10
conda activate ONLY
git clone https://github.com/zifuwan/ONLY.git
cd ONLY
pip install -r requirements.txt
python -m pip install -e transformers
```

Please also download the model checkpoints:

- [**LLaVA-1.5**](https://github.com/haotian-liu/LLaVA): Download [LLaVA-1.5 merged 7B](https://huggingface.co/liuhaotian/llava-v1.5-7b)
- [**InstructBLIP**](https://github.com/salesforce/LAVIS/tree/main/projects/instructblip): Download [InstructBLIP](https://huggingface.co/Salesforce/instructblip-vicuna-7b)
- [**Qwen-VL-Chat**](https://huggingface.co/Qwen/Qwen-VL-Chat): Download [Qwen-VL-Chat](https://huggingface.co/Qwen/Qwen-VL-Chat/tree/main)

As for the datasets and benchmarks:

- For **MSCOCO** dataset, see [this link](https://cocodataset.org/).
- For **MME**, see [this link](https://github.com/BradyFU/Awesome-Multimodal-Large-Language-Models/tree/Evaluation).

## 📦Usage

We provide the code for evaluating our ONLY on POPE, CHAIR, and MME-Hallucination benchmark. You can simply run the following code to run the experiments:

- POPE: `bash eval_bench/scripts/pope_eval.sh`
- CHAIR:`bash eval_bench/scripts/chair_eval.sh`
- MME:`bash experiments/cd_scripts/mme_eval.sh`

## 🙏Acknowledgements

Our codebase is adapted from  [RITUAL](https://github.com/sangminwoo/RITUAL), [VCD](https://github.com/DAMO-NLP-SG/VCD), [OPERA](https://github.com/shikiw/OPERA), [LLaVA](https://github.com/haotian-liu/LLaVA), and [DeGF](https://github.com/zhangce01/DeGF/tree/main). We thank the authors for releasing their code!

## 📧Contact

If you have any questions, please  contact [zifuw@andrew.cmu.edu](mailto:zifuw@andrew.cmu.edu).

## 📌 BibTeX & Citation

If you find this code useful, please consider citing our work:

```bibtex
@article{wan2025only,
  title={ONLY: One-Layer Intervention Sufficiently Mitigates Hallucinations in Large Vision-Language Models},
  author={Wan, Zifu and Zhang, Ce and Yong, Silong and Ma, Martin Q and Stepputtis, Simon and Morency, Louis-Philippe and Ramanan, Deva and Sycara, Katia and Xie, Yaqi},
  journal={arXiv preprint arXiv:2507.00898},
  year={2025}
}
```

ONLY

ONLY 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

ONLY

ONLY 최신버젼 다운로드

오늘의 추천 프로젝트

Was this content helpful?

Recent Feedback

Download README