BAN-Cap

BAN-Cap数据集是由Shahjalal University of Science and Technology的研究团队开发的，旨在解决孟加拉语在自动图像字幕生成领域缺乏标准数据集的问题。该数据集扩展自Flickr8k，包含8091张图片，总计40455对英文-孟加拉语字幕，由精通两种语言的孟加拉语母语者进行标注。数据集内容丰富，涵盖多种图像字幕风格，适用于多模态机器翻译、视觉问答等研究领域，特别强调在孟加拉语-英语和英语-孟加拉语翻译中的应用。创建过程中，通过严格的标注和后处理确保数据质量。该数据集不仅支持图像字幕生成，还展示了其在多用途研究中的潜力，如神经机器翻译和文本增强技术。

The BAN-Cap dataset was developed by a research team from Shahjalal University of Science and Technology, aiming to address the shortage of standard datasets for Bengali in the field of automatic image captioning. Expanded from Flickr8k, this dataset contains 8091 images and a total of 40455 English-Bengali caption pairs, which were annotated by Bengali native speakers proficient in both languages. The dataset features rich content covering various image captioning styles, and is applicable to research fields such as multimodal machine translation and visual question answering, with particular emphasis on its applications in Bengali-English and English-Bengali translation. During the creation process, strict annotation and post-processing steps were implemented to ensure data quality. This dataset not only supports image caption generation, but also demonstrates its potential in multi-purpose research such as neural machine translation and text augmentation techniques.