Image Copy Detection for Diffusion Models
Wenhao Wang, Yifan Sun, Zhentao Tan, Yi Yang
2024-10-01
Summary
This paper discusses ICDiff, a new model designed to detect when images generated by diffusion models copy or replicate existing images, addressing concerns about originality in digital artwork.
What's the problem?
As diffusion models become more popular for creating digital images, there's a growing concern that these models might unintentionally reproduce content from existing images. Current methods for detecting copied images work well for traditional hand-crafted replicas but struggle with detecting copies made by diffusion models.
What's the solution?
To tackle this issue, the authors developed ICDiff, the first image copy detection model specifically for diffusion-generated images. They created a new dataset called Diffusion-Replication (D-Rep), which includes 40,000 pairs of original and generated images labeled with different levels of replication. They introduced a method called PDF-Embedding that transforms the replication levels into a probability density function to help the model learn how to identify copied content more effectively.
Why it matters?
This research is important because it helps ensure the originality of digital content created by AI. By improving the ability to detect copied images, ICDiff can protect artists' rights and maintain the integrity of digital artwork in marketing and other fields.
Abstract
Images produced by diffusion models are increasingly popular in digital artwork and visual marketing. However, such generated images might replicate content from existing ones and pose the challenge of content originality. Existing Image Copy Detection (ICD) models, though accurate in detecting hand-crafted replicas, overlook the challenge from diffusion models. This motivates us to introduce ICDiff, the first ICD specialized for diffusion models. To this end, we construct a Diffusion-Replication (D-Rep) dataset and correspondingly propose a novel deep embedding method. D-Rep uses a state-of-the-art diffusion model (Stable Diffusion V1.5) to generate 40, 000 image-replica pairs, which are manually annotated into 6 replication levels ranging from 0 (no replication) to 5 (total replication). Our method, PDF-Embedding, transforms the replication level of each image-replica pair into a probability density function (PDF) as the supervision signal. The intuition is that the probability of neighboring replication levels should be continuous and smooth. Experimental results show that PDF-Embedding surpasses protocol-driven methods and non-PDF choices on the D-Rep test set. Moreover, by utilizing PDF-Embedding, we find that the replication ratios of well-known diffusion models against an open-source gallery range from 10% to 20%.