FlipBook: Exploring Domain Generalization via Diffusion-Based Interpolation.
Abi, Chaimaa
Abi, Chaimaa
Author
Supervisor
Department
Computer Vision
Embargo End Date
2025-05-30
Type
Thesis
Date
2025
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Domain generalization remains a fundamental challenge in computer vision, where models often struggle to maintain performance when faced with domain shifts not seen during training. While recent advances in vision-language models have shown promise in cross-domain transfer, they still face limitations in handling significant domain variations. This thesis addresses this challenge by developing a framework that combines multimodal generation with consistency-based learning for robust domain generalization. The key hypothesis of this work is that generating controlled domain variations through multi-modal models while enforcing consistency constraints can lead to more robust domain generalization. We propose that by leveraging the semantic understanding of large language models and the generative capabilities of diffusion models, we can create meaningful domain variations that preserve class-specific characteristics.
Citation
Chaimaa Abi, “FlipBook: Exploring Domain Generalization via Diffusion-Based Interpolation.,” Master of Science thesis, Computer Vision, MBZUAI, 2025.
Source
Conference
Keywords
Domain Generalization, Diffusion Models, Consistency Regularization