ManiFPT: Defining and Analyzing Fingerprints of Generative Models

TL;DR

ManiFPT defines GM artifacts and fingerprints by projecting generated samples onto an estimated real data manifold in multiple embedding spaces, with artifacts forming the model fingerprint F_G. The authors relate these fingerprints to Precision/Recall and IPMs, provide a practical estimation algorithm, and deploy a ResNet50-based attribution network to identify the source GM from artifact representations. Through extensive experiments on four comprehensive GM datasets spanning GANs, VAEs, Flows, and diffusion models, ManiFPT outperforms prior baselines in multi-class model attribution and demonstrates strong cross-dataset generalization. Analyses show learned artifact spaces offer superior separability and that fingerprint structure closely tracks design choices like upsampling and loss functions, underscoring the fingerprints' interpretability and practical utility for understanding and monitoring generative models.

Abstract

Recent works have shown that generative models leave traces of their underlying generative process on the generated samples, broadly referred to as fingerprints of a generative model, and have studied their utility in detecting synthetic images from real ones. However, the extend to which these fingerprints can distinguish between various types of synthetic image and help identify the underlying generative process remain under-explored. In particular, the very definition of a fingerprint remains unclear, to our knowledge. To that end, in this work, we formalize the definition of artifact and fingerprint in generative models, propose an algorithm for computing them in practice, and finally study its effectiveness in distinguishing a large array of different generative models. We find that using our proposed definition can significantly improve the performance on the task of identifying the underlying generative process from samples (model attribution) compared to existing methods. Additionally, we study the structure of the fingerprints, and observe that it is very predictive of the effect of different design choices on the generative process.

Figures (9)

• Figure 1: Our definition of artifacts and fingerprints of a generative model. We estimate the true data manifold $\mathcal{M}$ using real samples and compute an artifact $a$ as the difference between a generated sample and its closest point in the real dataset. We define the fingerprint F of a generative model as the set of all its artifacts.
• Figure 2: Features learnt using our definition of artifacts (f) achieve better separation between samples from different generative models (shown in different colors). (a) Shows tSNE of generated samples in pixel space, (b) in the latent space of ResNet50 pretrained on ResNet50, (c-f) in the penultimate layer of the classifier proposed by each method trained on the task of model attribution.
• Figure 3: Our attribution method. We propose a model attribution method based on our definition of artifact as deviations from an estimate data manifold. Given input images $X_G$, we first map the images to a chosen embedding space (RGB, Frequency, a feature space of a pretrained supervised-learning (SL) or self-supervised leanring (SSL) network) and compute their artifacts $a$. We then pass the artifacts to a ResNet50-based attribution network (Model Attributor) and fine-tune the network to identify the source generative model under the (multi-class) cross-entropy loss.
• Figure 4: We visualize artifacts in generated images under our manifold-based definition (Sec. \ref{['defn:artifact']}). Each row shows an original image generated by a generative model, followed by its projection to data manifolds in RGB ($x_{\text{RGB}}^{\star}$), Frequency ($x_{\text{FREQ}}^{\star}$), and learned feature spaces of SL ($x_{\text{SL}}^{\star}$) and SSL ($x_{\text{SSL}}^{\star}$). The third and fourth columns show our definition of artifacts in the RGB and frequency spaces, respectively. Note that artifacts in SL and SSL spaces are not shown as they are 2048-long vectors (in the embedding space of a pretrained ResNet50).
• Figure 5: Visualization of artifacts in the RGB space (GM-CelebA). Each column corresponds to the generated images ($x_G$), their closest points on the data manifold ($x^{\star}$), and the artifacts ($a$). Each artifact is computed as the different between $x^{\star}$ and $x_G$ following the definition and algorithm in Sec. \ref{['sec:our-defns']}.

Theorems & Definitions (2)

• Definition 3.1: Artifact
• Definition 3.2: Fingerprint