Text To Image Model Using AI

Abstract

This report presents a project that explores the generation of images from textual descriptions using AI and Generative Adversarial Networks (GANs). The project focuses on utilizing the Oxford Flowers 102 dataset, which consists of 102 categories of flower images. By harnessing the power of deep learning techniques and GANs, the aim is to develop a model capable of generating realistic and visually appealing flower images based solely on textual input. The project begins with preprocessing the dataset, including image resizing and text encoding, to prepare it for training the GAN model. A GAN architecture is then designed and trained using an adversarial learning framework, with the generator network responsible for generating images from text and the discriminator network learning to differentiate between real and generated images. To enhance the quality of the generated images, various techniques such as conditional GANs and attention mechanisms are explored. The conditional GANs allow the model to condition the image generation process on the given text, enabling the generation of more specific and accurate images. Attention mechanisms aid in focusing the model’s attention on relevant parts of the text during the image generation process, further improving the fidelity and coherence of the generated images. Evaluation of the model’s performance is conducted through quantitative and qualitative analysis. Quantitatively, metrics such as Fréchet Inception Distance (FID) and Inception Score (IS) are employed to assess the quality and diversity of the generated images. Qualitative evaluation involves human judgment and comparison with ground truth images from the Oxford Flowers 102 dataset. The experimental results demonstrate the effectiveness of the proposed model in generating visually plausible flower images from textual descriptions. The generated images exhibit fine-grained details and capture the essence of the described flowers. The research outcomes not only contribute to the field of image synthesis but also have potential applications in areas such as computer-aided design, virtual reality, and creative arts.