moranyanuka/OpenCHAIR
收藏资源简介:
OpenCHAIR数据集是一个用于评估视觉语言模型(VLMs)在生成描述时产生幻觉问题的数据集。该数据集包含图像和文本特征,主要用于研究如何减少开放词汇描述中的幻觉现象。数据集由测试集组成,包含2000个样本,总大小为309448153字节。该数据集在论文《MOCHa: Multi-Objective Reinforcement Mitigating Caption Hallucinations》中首次引入,并提供了相关的GitHub代码库用于评估。
The OpenCHAIR dataset is a benchmark for evaluating hallucination issues in caption generation for Vision-Language Models (VLMs). It includes image and text features, and is primarily used to study methods for reducing hallucinations in open-vocabulary captioning. The dataset consists of a test split with 2000 samples, and has a total size of 309,448,153 bytes. This dataset was first introduced in the paper *MOCHa: Multi-Objective Reinforcement Mitigating Caption Hallucinations*, and a corresponding GitHub repository for evaluation is provided.
OpenCHAIR 数据集概述
数据集名称
- OpenCHAIR
数据集介绍
- 该数据集是在“MOCHa: Multi-Objective Reinforcement Mitigating Caption Hallucinations”中正式引入的。




