Guided Image Synthesis

Guided image synthesis aims to generate new images based on given guidance, such as text descriptions, sketches, or reference images, achieving high fidelity and control over the generated content. Current research heavily utilizes diffusion models, often incorporating techniques like classifier-free guidance and progressive refinement stages to improve image quality and control over specific features. This field is significant for its applications in various domains, including medical image analysis, vehicle re-identification, and data augmentation for improving the performance of downstream tasks like image segmentation and classification.