Machine Learning enables near-perfect algorithmic compositions. The proposed solution, Stacked Generative Adversarial Networks, generates photo-realistic images from text descriptions by decomposing the problem into manageable sub-problems through a sketch-refinement process. The Stage-I GAN sketches low-resolution images of the object's primitive shape and colors. The Stage-II GAN generates high-resolution images with photo-realistic details by rectifying defects in Stage-I results and adding compelling details with the refinement process. A Conditioning Augmentation technique improves diversity and stabilizes training. The proposed method achieves significant improvements in generating photo-realistic images conditioned on text descriptions.
Article Details
Unique Paper ID: 159511
Publication Volume & Issue: Volume 9, Issue 12
Page(s): 1065 - 1069
Article Preview & Download
Share This Article
Join our RMS
Conference Alert
NCSEM 2024
National Conference on Sustainable Engineering and Management - 2024