Single-shot image generation and stylisation via cross-domain correspondance

• Main Author: Kalyan, Harikishan
• Other Authors: Lin Guosheng
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2022
• Subjects: Engineering::Computer science and engineering
• Online Access: https://hdl.handle.net/10356/163314
• Institution: Nanyang Technological University

The advent of generative adversarial networks has led to many state-of-the-art methodologies in the field of image generation and stylisation. Among the most popular methods to generate new and diverse images is cross-domain correspondence, where the generated output would be a mix of the stylistic elements and attributes of a source dataset and a target dataset. This method, however, can be resource intensive due to the need for massive datasets. Existing methodologies from Ojha et al and Mind the Gap have attempted to address this issue by requiring only a few images for domain adaptation, they are prone to overfitting concerns due to a limited dataset. To counter these problems, a CLIP guided domain adaptation approach is proposed where only a single image is needed for the model to generate diverse images of various styles