Image Composition
Image composition focuses on seamlessly integrating objects into new scenes, aiming for realistic and coherent results. Current research emphasizes training-free methods leveraging pre-trained diffusion models, often incorporating techniques like latent space manipulation, attention mechanisms, and depth map utilization to improve object placement, occlusion handling, and overall visual fidelity. This area is significant for advancements in image editing, generation, and retrieval, with applications ranging from artistic creation and video game development to historical image analysis and data augmentation.
Papers
Learning Visual Composition through Improved Semantic Guidance
Austin Stone, Hagen Soltau, Robert Geirhos, Xi Yi, Ye Xia, Bingyi Cao, Kaifeng Chen, Abhijit Ogale, Jonathon Shlens
Affordance-Aware Object Insertion via Mask-Aware Dual Diffusion
Jixuan He, Wanhua Li, Ye Liu, Junsik Kim, Donglai Wei, Hanspeter Pfister