Conditional Transformation Diffusion for Object Placement
Jiacheng Liu, Shida Wei, Rui Ma · 2024
We propose a conditional transformation diffusion model for the object placement task in the field of image composition. Specifically, we design an object and background feature extractor (OBFE) to infuse information about object and background into the diffusion model, enhancing its ability in object placement. Our proposed model has the advantages of having fewer parameters, being easy to train, producing high-quality and diverse composite images. Through experiments, we demonstrate that on the OPA dataset, compared to other GAN-based methods, our approach achieves the best results in terms of both plausibility and diversity.