Semantic prediction speeds object removal in images to one step
SERO: Semantic Prediction for Single-Step Object Removal
Computer Vision and Pattern Recognition
Summary
Removing objects from pictures and filling in the empty space is tricky because there are many ways to do it that look right but may not fit the scene’s meaning. The authors present a new method called SERO that predicts what the missing area should contain based on the scene, then uses that to complete the image in a single step. This makes the result both visually realistic and consistent with the surroundings, while also being much faster than older methods that needed many steps. Tests show SERO works as well or better than existing tools on benchmark image sets.
What this means in practice
- •For photo editing software developers: Implement fast, one-step object removal that produces contextually consistent image completions.$Commercial implications: Enables next-generation photo editors to offer high-quality object removal with minimal delay, improving user experience in commercial apps.
- •For game developers: Accelerate scene retouching by quickly removing objects and filling in backgrounds plausibly in virtual environments.
Authors
Amreen Hasheem Siraj, Alireza Javanmardi, Pragati Jaiswal, Didier Stricker, Alain Pagani
Abstract
Object removal and semantic inpainting of missing regions remain inherently ill-posed, as multiple plausible completions may exist for a given scene. Recent diffusion-based methods have demonstrated strong potential for high-quality object removal; however, they still face two key limitations. First, although they can synthesise sharp and high-resolution images, they often struggle to complete removed regions in a semantically consistent and contextually meaningful manner, where the generated content should agree with the surrounding scene layout and object relationships rather than merely appear visually realistic. Second, their iterative denoising process makes them computationally expensive, limiting their practicality for downstream applications. We propose SERO, a semantic prediction framework for single-step object removal. Our model predicts a hidden-region semantic prior in embedding space, which guides a latent bridge-matching generator toward plausible scene completion by providing both where to edit and what structure is likely to exist behind the removed object. This enables visually realistic and contextually coherent completions in one forward pass. Experiments on the RORD and OBER datasets show that SERO achieves competitive or state-of-the-art performance while requiring only a single generation step.