Bidirectional Vision Language Translation Using CNN & RNN

Abstract

 This paper presents Bidirectional vision-language translation enables conversion between images and text, bridging visual and linguistic modalities. This work presents a deep learning framework combining Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) for both image-to-text and text-to-image tasks. In the forward direction, CNNs extract visual features that are decoded by an RNN to generate descriptive captions. In the reverse direction, textual inputs are encoded using an RNN, and a generative model reconstructs corresponding images. Trained on paired image-caption datasets, the model learns shared cross-modal representations for coherent and context-aware translation. Experimental results demonstrate effective caption generation and visually relevant image synthesis. The proposed approach highlights the potential of CNN–RNN architectures for multimodal applications such as assistive systems, content generation, and human-computer interaction.

DOI: https://doi-ds.org/doilink/04.2026-33381126