ArtEmis v2.0

该数据集是一个大规模的情绪反应图像集，其中包含了对这些情绪的语言解释，为了减少情绪偏差，还引入了一个新的补充数据集。该数据集共包含260,533个实例，旨在提高情绪多样性并减少在情绪表达上的偏见。这项任务的目的是进行情感图像字幕生成和情绪分析。

This is a large-scale image dataset of emotional reactions, paired with linguistic explanations for the depicted emotions. To mitigate emotional bias, an additional supplementary dataset is introduced. Collectively, these datasets contain 260,533 instances in total, with the objectives of enhancing emotional diversity and reducing biases in emotional expression. The objective of this task is to perform emotional image captioning and emotion analysis.

搜集汇总

背景与挑战

背景概述

ArtEmis v2.0是一个大规模的情感反应与语言解释数据集，通过对比数据收集方法平衡了情感分布，包含260,533个新增实例，显著提升了神经网络在情感分类和图像描述任务上的性能。