AIML-TUDA/t2i-diversity-gender-neutral-captions

这个数据集包含了用于图像样本的不同合成标题。通过实验选择了表现最佳的随机长度标题集合，并从中移除了性别指示性词汇，生成了三种不同的标题集合：性别完全中性化、仅女性中性化、仅男性中性化。这样做的目的是为了研究性别缺失或存在如何影响下游任务的性别偏倚。This dataset contains different synthetic captions for image samples. The best-performing caption set from our experiments, the random-length captions, has been gender-neutralized in three different ways: fully neutralized, female-only neutralized, and male-only neutralized. The purpose of this is to experiment with how the absence or presence of gender(s) affects downstream gender bias.