GD-VCR

该数据集旨在测试视觉和语言模型理解文化及地理位置特定常识知识的能力。它评估了模型在需要特定于不同地区的常识推理问题上的泛化能力，并突显了在西方和非西方情境下模型性能的差距。该数据集的任务是视觉常识推理。

This dataset is designed to test the ability of vision-language models to comprehend culturally and geographically specific commonsense knowledge. It evaluates the generalization capability of models when dealing with commonsense reasoning questions that require region-specific knowledge, and highlights the performance gaps between models in Western and non-Western contexts. The task of this dataset is visual commonsense reasoning.

搜集汇总

背景与挑战

背景概述

GD-VCR是一个视觉常识推理数据集，专注于评估模型在文化和地理位置特定常识知识上的理解能力。它旨在测试模型在不同地区常识推理问题中的泛化表现，并揭示西方与非西方情境下模型性能的差异。