Hybrid Generative-Discriminative Object Placement
2026-08-24 • Computer Vision and Pattern Recognition
Computer Vision and Pattern Recognition
AI summaryⓘ
The authors focus on placing objects into images in a way that looks natural, figuring out the best spot and size for the object. Their method is a mix between two common approaches, aiming to be both fast and accurate. They place multiple possible positions uniformly over the background and then score each to find the most plausible spots. Tests show their method works well balancing speed and quality.
object placementimage compositionforegroundbackgroundgenerative methodsdiscriminative methodssemi-generative methodanchor pointsrationality scoreOPA dataset
Authors
Siyuan Zhou, Li Niu
Abstract
As an important operation of image composition, object placement aims to predict the plausible placement (location, scale) for the inserted foreground object. Previous object placement methods can be divided into generative methods and discriminative methods, both of which cannot balance efficiency and effectiveness well. In this work, we propose a semi-generative method in the middle ground between them. In particular, we assign uniformly distributed anchors on the background. Then, we fuse foreground and background features to predict the rationality score for each anchor and predict plausible placement sets for positive anchors. Extensive experiments on the OPA dataset show that our method can strike a good balance between efficiency and effectiveness.