LU-500: A Logo Benchmark for Concept Unlearning

2026-07-27

AI Summary

This work addresses the challenge of effectively erasing localized, semantically entangled corporate logos—a task poorly handled by existing concept erasure methods and lacking dedicated evaluation benchmarks. To bridge this gap, we introduce LU-500, a benchmark comprising nearly 10,000 text–logo image pairs with both explicit and implicit triggers, along with a multi-granularity, spatially aware evaluation protocol that jointly assesses local logo removal accuracy and global image fidelity. Focusing for the first time on localized, semantically nested logo forgetting, we systematically evaluate inference-time and fine-tuning-based approaches (e.g., NP, ESD) alongside a prompt-space multi-agent baseline, ProLU. Experiments reveal that current methods often damage non-target content while attempting precise logo removal. Further analysis shows that logo area, position, and structural complexity significantly impact forgetting efficacy, suggesting that spatially aware constraints—such as SSIM-guided optimization—are essential for improved performance.

Abstract

Concept unlearning is increasingly used to limit the reproduction of protected or unsafe visual concepts in text-to-image models. Existing evaluations, however, mostly study targets that dominate the whole image, such as styles, broad object categories, or portrait-like identities, leaving company logos comparatively underexamined. Logos create a different failure mode: a small localized mark can carry the entire protected concept, must be visually precise to remain recognizable, and can be triggered implicitly by products, storefronts, packaging, or advertisements even when the word ``logo'' is absent. We introduce LU-500, a logo-unlearning benchmark built from Fortune Global 500 companies to study this localized and semantically entangled setting. LU-500 contains nearly 10,000 curated text-query and logo-image pairs, with an explicit track (LUex-500) and an implicit contextual track (LUim-500). To avoid reducing the task to a binary detector score, we define a multi-grained protocol that evaluates both local logo removal and global image preservation in pixel and latent spaces. Experiments on representative inference-time methods, including NP, SLD, and SEGA, and compatible fine-tuning-based methods such as ESD and Forget-Me-Not, show that the evaluated methods struggle to remove logo evidence without changing non-target content. We further analyze ProLU, a prompt-space multi-agent baseline: it improves local erasure by removing logo-inducing semantics, but also illustrates why prompt filtering is not a substitute for weight-level disentanglement. Correlation analyses over logo area, location, and structural complexity suggest that future logo unlearning may need spatially aware controls, such as SSIM-guided constraints, rather than purely global concept suppression.

Problem

• concept unlearning
• logo removal
• text-to-image generation
• localized visual concepts
• semantic entanglement

Innovation

• concept unlearning
• logo benchmark
• localized visual concepts
• multi-grained evaluation
• spatially aware control