A Dataset and Benchmark for Copyright Infringement Unlearning from Text-to-Image Diffusion Models

TL;DR

Copyright infringement risks in text-to-image diffusion models motivate the work. The authors assemble the CPDM dataset and benchmark by connecting anchor copyright images, prompts derived via CLIP interrogation, and diffusion-generated infringing images in a unified pipeline. They define the CPDM metric, CM, as $CM = 1 - Norm(M_sem × M_style)^2$, combining a semantic component from CLIP embeddings with a Gram-matrix-based style component, and they evaluate unlearning methods to curb infringement while preserving non-infringing generation. The results, supported by human verification, demonstrate a meaningful, publicly available resource to drive research and practical safeguards in copyright protection for AIGC.

Abstract

Copyright law confers upon creators the exclusive rights to reproduce, distribute, and monetize their creative works. However, recent progress in text-to-image generation has introduced formidable challenges to copyright enforcement. These technologies enable the unauthorized learning and replication of copyrighted content, artistic creations, and likenesses, leading to the proliferation of unregulated content. Notably, models like stable diffusion, which excel in text-to-image synthesis, heighten the risk of copyright infringement and unauthorized distribution.Machine unlearning, which seeks to eradicate the influence of specific data or concepts from machine learning models, emerges as a promising solution by eliminating the \enquote{copyright memories} ingrained in diffusion models. Yet, the absence of comprehensive large-scale datasets and standardized benchmarks for evaluating the efficacy of unlearning techniques in the copyright protection scenarios impedes the development of more effective unlearning methods. To address this gap, we introduce a novel pipeline that harmonizes CLIP, ChatGPT, and diffusion models to curate a dataset. This dataset encompasses anchor images, associated prompts, and images synthesized by text-to-image models. Additionally, we have developed a mixed metric based on semantic and style information, validated through both human and artist assessments, to gauge the effectiveness of unlearning approaches. Our dataset, benchmark library, and evaluation metrics will be made publicly available to foster future research and practical applications (https://rmpku.github.io/CPDM-page/, website / http://149.104.22.83/unlearning.tar.gz, dataset).

Figures (7)

• Figure 1: The pipeline of Copyright Unlearning and the crucial role of Datasets and Benchmarks.
• Figure 2: Examples of CPDM dataset composition and unlearned results for copyright protection.
• Figure 3: Pipeline for the CPDM Dataset Creation. The clip-interrogator is utilized to convert copyrighted images into corresponding textual information. This text is subsequently refined and transformed into prompts, which are then inputted into a diffusion model to generate the corresponding infringing images.
• Figure 4: Analysis of Dataset Style Diversity. We conducted a statistical of the styles in prompts obtained using the pipeline in Fig. \ref{['fig:prompt_pipeline']}. The various styles at the bottom are provided by artists.
• Figure 5: Analysis and Comparison of the Effectiveness of Evaluation Metrics.