The Text-to-Image Synthesis project aims to develop a robust deep learning model capable of generating realistic images from textual descriptions. This cutting-edge application bridges the gap between natural language understanding and image synthesis, providing a valuable tool for various industries such as e-commerce, graphic design, and virtual/augmented reality.
- Name: Flickr 30k
- Description: This dataset consists of 30,000 images collected from Flickr, each with five descriptive sentences provided by human annotators.
StackGAN utilizes Stacked Generative Adversarial Networks (GANs) to generate high-resolution images with photo-realistic details. The generative process is decomposed into two stages:
- Stage-I GAN: Sketches the primitive shape and basic colors of the object based on the given text description. It draws the background layout from a random noise vector, resulting in a low-resolution image.
- Stage-II GAN: Corrects defects in the low-resolution image from Stage-I and completes the details of the object by revisiting the text description. It produces a high-resolution, photo-realistic image.
Figure: StackGAN Architecture
- Earlier architectures faced limitations in generating high-resolution and diverse images from text descriptions.
- StackGAN employs a multi-stage generation process, capturing both coarse and fine-grained details for more realistic images.
- It effectively encodes textual descriptions into a format usable by the generator network, resulting in more coherent image generation compared to earlier methods.
- Conditional Image Generation with PixelCNN Decoders (PixelCNN) - 2016
- Generative Adversarial Text-to-Image Synthesis (GAN-INT-CLS) - 2016
- StackGAN: Text to Photo-realistic Image Synthesis with Stacked Generative Adversarial Networks - 2017
- Clone the repository: https://github.com/AnruthaKamal/Text-Canvas-.git
- Install dependencies: pip install -r requirements.txt