Compositional EffectsSource / Target / Mask
Real-world video object removal
EffectLearner
World-Aware Object-Effect Reasoning for Real-World Video Object Removal
Paper overview
Abstract
Video object removal must eliminate not only the target object but also its induced effects while maintaining high-fidelity and spatiotemporally coherent restoration. Existing methods mainly learn object-effect correspondences implicitly from predefined effect categories and fixed data distributions, limiting their generalization to complex real-world scenes involving compositional effects, spatially detached or weakly correlated effects, long-tail physical phenomena, and dynamically evolving interactions.
We propose EffectLearner, a semantic-reasoning-enhanced framework that combines a VLM-based Object-Effect Reasoner with a DiT-based Video Eraser. Guided by a structured effect-analysis prompt, the Reasoner performs cross-modal reasoning over a target-highlighted video and extracts compact effect-aware context, which guides the Video Eraser toward comprehensive object-effect removal. Motion-aware mask guidance and motion-consistency supervision further improve removal coverage and spatiotemporal stability under object motion and evolving scene dynamics.
To fully exploit the framework in challenging real-world scenarios, we further construct EffectWorld, a paired video dataset specifically designed for complex object-induced effects, and introduce a progressive training curriculum that combines conventional supervision with complex-effect data. On the standard ROSE-Bench, EffectLearner outperforms existing baselines on most metrics and achieves clear advantages on both EffectWorld-Eval and the challenging EffectWorld-Wild, demonstrating its ability to deliver high-quality video object removal in complex real-world scenes.
Qualitative Comparison
Object-and-effect removal
Four case rows per page with one fixed model per column. All videos are synchronized over the same 49-frame, 15 fps temporal range.
1 / 1
Dataset Demo
Complex-effect triplets and target masks
Examples are organized by the same four challenge categories used throughout the paper and project page.
01
Compositional Effects
Multiple object-induced effects that must be removed together.
Compositional EffectsSource / Target / Mask
02
Weak Object-Effect Correlation
Spatially detached or weakly connected effects beyond the target support.
Weak Object-Effect CorrelationSource / Target / Mask
Weak Object-Effect CorrelationSource / Target / Mask
03
Long-Tail Physical Phenomena
Physical traces that extend beyond conventional side-effect categories.
Long-Tail Physical PhenomenaSource / Target / Mask
Long-Tail Physical PhenomenaSource / Target / Mask
Data Preparation Pipeline
EffectWorld construction
EffectWorld combines compositional effects, weak object-effect correlations, long-tail physical phenomena, and dynamic challenges. Controlled object and camera trajectories produce spatially and temporally aligned source, target, and mask sequences.
Method
EffectLearner framework
The VLM-based Object-Effect Reasoner extracts compact semantic context from a target-highlighted video and structured prompt. The DiT-based Video Eraser combines this context with visual conditions, while motion-aware mask guidance and motion-consistency supervision stabilize dynamic restoration.
Reference
BibTeX
@misc{wu2026effectlearnerworldawareobjecteffectreasoning,
title={EffectLearner: World-Aware Object-Effect Reasoning for Real-World Video Object Removal},
author={Feier Wu and Wanke Xia and Xu He and Zilang Zhou and Si Chen and Dongxia Liu and Liyang Chen and Qimeng Wu and Zhengbo Zhang and Wenming Yang and Zhiyong Wu},
year={2026},
eprint={2608.05565},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2608.05565}
}