ICML Workshops 26 · 2026

Images as Tables - In-Context Learning with TabPFN for Low-Data Detection of AI-Generated Images

Jan Philip Walter, Shashank Agnihotri, Margret Keuper

Why this publication matters

A detector trained on yesterday's image generator may struggle with tomorrow's. This method combines fixed image features with a model that learns from a small set of examples provided as context. It offers a lightweight way to adapt detection without repeatedly retraining a large vision network.

Abstract

AI-generated image detection is a moving-target problem: detectors trained on one generator often fail when a new generator appears, and only a few labeled examples are available. We study a simple image-to-table formulation for this regime, where each image is encoded by a frozen DINOv3 backbone, its CLS feature is reduced to a 500-dimensional structured row with PCA, and TabPFN performs real/fake classification by incontext tabular inference rather than task-specific classifier training. This turns fake-image detection into low-data structured prediction over learned visual features, making detector adaptation depend on the labeled context set instead of gradient-based fine-tuning. On GenImage, LATTE, a recent state-of-the-art detector, remains stronger when many labeled samples from all generators are available, by 7.4% in the largest pooled setting, but DINOv3-PCA-TabPFN is stronger in the practically important low-data regime, outperforming LATTE by up to 8.2%, and in transfer settings where the detector must generalize from one generator to another. These results position tabular foundation models as a strong complementary adaptation mechanism for image forensics, shifting adaptation from detector retraining to lightweight in-context updates with a small labeled set of examples. Code hyperlink here.

Abstract source ↗

Figures

AI-image detection accuracy as the number of labeled examples increases. — **Figure 1.** In the pooled generator setting, LATTE reaches the highest accuracy (real/fake class detection) at the largest training size, but DINOv3-PCA-TabPFN is stronger at the smaller shared training sizes, which is the regime targeted by our in-context detector adaptation. View in source ↗

**Figure 2.** Representation and classifier comparison in the Multi-Multi development setting. DINOv3 features with TabPFN give the strongest and most stable performance across accuracy, precision, recall, F1, and ROC-AUC, while DINOv2, frequency features, and the MLP baseline are weaker in the low-data regime. All compared encoders and settings are explained in Appendix Section C. View in source ↗

Cite this paper

Download .bib

@inproceedings{walter2026imagesastables82,
  title = {{Images as Tables - In-Context Learning with TabPFN for Low-Data Detection of AI-Generated Images}},
  author = {Jan Philip Walter and Shashank Agnihotri and Margret Keuper},
  booktitle = {ICML Workshops 26},
  year = {2026},
  url = {https://arxiv.org/pdf/2606.00872}
}

Figures and abstract are reproduced from the linked research sources. Credit remains with the authors and publishers.