Scholarlink

Images are integral to our digital experiences, and combining visual elements with verbal storytelling is crucial. While English image captioning has progressed significantly, Arabic image captioning lags behind, with limited research on Arabic story generation from images. This paper contributes by developing the first approach to generate an Arabic story from an image. It aims to enhance Arabic image captioning, an essential step in image-to-story generation, using advanced generated captions. Four Arabic image captioning models were implemented: an inject model, two merge models, and a model using a pre-trained vision transformer with AraGPT. These models used various vision transformers, CNNs, and Arabic pre-trained embedding models, trained on the Arabic flicker8k dataset. Captions generated from the best version of each model were input into GPT-3.5 Turbo for story generation.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Visualizing the Unseen: Arabic Image-to-Story Generation Using Deep Learning Techniques

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Eman Saleh,
  • Caroline Sabty

摘要

Images are integral to our digital experiences, and combining visual elements with verbal storytelling is crucial. While English image captioning has progressed significantly, Arabic image captioning lags behind, with limited research on Arabic story generation from images. This paper contributes by developing the first approach to generate an Arabic story from an image. It aims to enhance Arabic image captioning, an essential step in image-to-story generation, using advanced generated captions. Four Arabic image captioning models were implemented: an inject model, two merge models, and a model using a pre-trained vision transformer with AraGPT. These models used various vision transformers, CNNs, and Arabic pre-trained embedding models, trained on the Arabic flicker8k dataset. Captions generated from the best version of each model were input into GPT-3.5 Turbo for story generation.