ACE-LoRA: Adaptive Orthogonal Decoupling for Continual Image Editing

TL;DR

ACE-LoRA introduces a novel continual learning framework for image editing that mitigates forgetting and scales effectively, supported by a new benchmark for evaluation.

Contribution

The paper proposes ACE-LoRA, a dynamic regularization method with a new benchmark, CIE-Bench, for continual image editing tasks.

Findings

ACE-LoRA outperforms existing methods in instruction fidelity and visual realism.

CIE-Bench provides a comprehensive evaluation protocol for continual image editing.

ACE-LoRA demonstrates robustness to catastrophic forgetting in experiments.

Abstract

State-of-the-art diffusion models often rely on parameter-efficient fine-tuning to perform specialized image editing tasks. However, real-world applications require continual adaptation to new tasks while preserving previously learned knowledge. Despite the practical necessity, continual learning for image editing remains largely underexplored. We propose ACE-LoRA, a dynamic regularization framework for continual image editing that effectively mitigates catastrophic forgetting. ACE-LoRA leverages Adaptive Orthogonal Decoupling to identify and orthogonalize task interference, and introduces a Rank-Invariant Historical Information Compression strategy to address scalability issues in continual updates. To facilitate continual learning in image editing and provide a standardized evaluation protocol, we introduce CIE-Bench, the first comprehensive benchmark in this domain. CIE-Bench encompasses diverse and practically relevant image editing scenarios with a balanced level of difficulty to effectively expose limitations of existing models while remaining compatible with parameter-efficient fine-tuning. Extensive experiments demonstrate that our method consistently outperforms existing baselines in terms of instruction fidelity, visual realism, and robustness to forgetting, establishing a strong foundation for continual learning in image editing.