BanglaLekhaImageCaptions

该数据集由孟加拉语中的图像和注释组成。这些图像是由两名母语为孟加拉语的成年人用孟加拉语注释的。所有流行的图像字幕数据集都以英语注释为主要的西方文化偏见。使用此类数据集来训练图像字幕系统，假设存在良好的英语到目标语言的翻译系统，并且原始数据集具有目标文化的元素。这两个假设都是错误的，导致需要在孟加拉语中建立与文化相关的数据集，以生成与孟加拉国和更广泛的次大陆背景相关的图像的适当图像标题。呈现的数据集由9,154图像组成。

This dataset comprises images and their annotations in Bengali. All images were annotated in Bengali by two adult native Bengali speakers. Most popular image captioning datasets primarily employ English annotations, which embed strong Western cultural biases. Training image captioning systems using such datasets relies on two flawed assumptions: first, that high-quality English-to-target-language translation systems are available, and second, that the original dataset contains elements of the target culture. Both of these assumptions are invalid, creating a critical need for culturally relevant Bengali datasets to generate appropriate image captions for images tied to Bangladesh and the broader South Asian subcontinental context. The presented dataset consists of 9,154 images.