PatchHead: Learning Spatial Patch Evidence for Generalizable AI-Generated Image Detection

• URL: http://arxiv.org/abs/2608.09223v1
• Date: Mon, 10 Aug 2026 07:47:22 GMT
• Title: PatchHead: Learning Spatial Patch Evidence for Generalizable AI-Generated Image Detection
• Abstract summary: We introduce PatchHead, a lightweight spatial aggregation head that preserves the two-dimensional organization of DINO patch tokens and integrates evidence across neighboring regions.<n>Across nine cross-dataset benchmarks spanning manually curated and in-the-wild settings, PatchHead ranks first on seven datasets and second on the remaining two.<n>It improves the strongest prior method from 91.6% to 94.6% in average balanced accuracy (+3.0 points) and raises the worst-case accuracy from 82.4% to 89.4% (+6.9 points)
• Score: 6.466210485064843
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: AI-generated image detectors generalize poorly when their training and test images originate from different generators or datasets. Despite the rich spatial representations produced by vision foundation models like DINO, existing detectors typically classify images using only the globally aggregated CLS token. We hypothesize that globally aggregating DINO features into a single CLS token obscures spatially distributed generation traces. To test this hypothesis, we introduce PatchHead, a lightweight spatial aggregation head that preserves the two-dimensional organization of DINO patch tokens and integrates evidence across neighboring regions. During training, we freeze the pretrained DINO backbone and optimize only the inserted LoRA adapters, PatchHead, and auxiliary projection head. Across nine cross-dataset benchmarks spanning manually curated and in-the-wild settings, PatchHead ranks first on seven datasets and second on the remaining two. It improves the strongest prior method from 91.6% to 94.6% in average balanced accuracy (+3.0 points) and raises the worst-case accuracy from 82.4% to 89.4% (+6.9 points), while introducing only 8.6% more trainable parameters and 0.08% additional FLOPs. Further qualitative analysis suggests that PatchHead (i) reduces class-conditional domain discrepancy, and (ii) redirects the representation from content-dominated saliency toward spatially distributed authenticity evidence. Together, these observations provide a representation-level account of why spatial patch aggregation transfers more reliably across generators and datasets than a single CLS-based global representation. Our code and models will be made available upon acceptance.

Related papers

• E2S-Pruner: Progressive Two-Stage Evidence Fusion for Visual Token Pruning in Vision-Language Models [12.765837057758127]
  We propose a two-stage evidence-fusion framework for visual token pruning.<n>E2S-Pruner represents each visual token using three states: important, unimportant, and uncertain.<n>On LLaVA-1.5-7B, E2S-Pruner retains 98.0%, 96.8%, and 90.6% of the aggregate performance when the average numbers of retained visual tokens are 192, 128, and 64.
  arXiv  Detail & Related papers  (2026-08-24T13:44:41Z)
• iSAGE: A Human-in-the-Loop Framework for Remote Sensing Semantic Segmentation via Sparse Point Supervision [0.1749935196721634]
  Existing human-in-the-loop frameworks expand sparse clicks into dense supervision via auxiliary machinery.<n>This paper hypothesizes that expert clicks targeting confident model errors, not arbitrary pixels, suffice to match dense supervision.<n>iSAGE is the only iterative human-in-the-loop framework operating without auxiliary machinery.
  arXiv  Detail & Related papers  (2026-06-08T20:09:35Z)
• Function-Space Decoupled Diffusion for Forward and Inverse Modeling in Carbon Capture and Storage [65.51149575007149]
  We present Fun-DDPS, a generative framework that combines function-space diffusion models with differentiable neural operator surrogates for both forward and inverse modeling.<n>Fun-DDPS produces physically consistent realizations free from the high-frequency artifacts observed in joint-state baselines.
  arXiv  Detail & Related papers  (2026-02-12T18:58:12Z)
• What matters for Representation Alignment: Global Information or Spatial Structure? [64.67092609921816]
  Representation alignment (REPA) guides generative training by distilling representations from a strong, pretrained vision encoder to intermediate diffusion features.<n>We investigate a fundamental question: what aspect of the target representation matters for generation, its textitglobal revisionsemantic information.<n>We replace the standard projection layer in REPA with a simple convolution layer and introduce a spatial normalization layer for the external representation.
  arXiv  Detail & Related papers  (2025-12-11T16:39:53Z)
• CLUE: Non-parametric Verification from Experience via Hidden-State Clustering [64.50919789875233]
  We show that correctness of a solution is encoded as a geometrically separable signature within the trajectory of hidden activations.<n>ClUE consistently outperforms LLM-as-a-judge baselines and matches or exceeds modern confidence-based methods in reranking candidates.
  arXiv  Detail & Related papers  (2025-10-02T02:14:33Z)
• Improved Anomaly Detection through Conditional Latent Space VAE Ensembles [49.1574468325115]
  Conditional Latent space Variational Autoencoder (CL-VAE) improved pre-processing for anomaly detection on data with known inlier classes and unknown outlier classes. Model shows increased accuracy in anomaly detection, achieving an AUC of 97.4% on the MNIST dataset. In addition, the CL-VAE shows increased benefits from ensembling, a more interpretable latent space, and an increased ability to learn patterns in complex data with limited model sizes.
  arXiv  Detail & Related papers  (2024-10-16T07:48:53Z)
• Less is More: Fewer Interpretable Region via Submodular Subset Selection [54.07758302264416]
  This paper re-models the above image attribution problem as a submodular subset selection problem. We construct a novel submodular function to discover more accurate small interpretation regions. For correctly predicted samples, the proposed method improves the Deletion and Insertion scores with an average of 4.9% and 2.5% gain relative to HSIC-Attribution.
  arXiv  Detail & Related papers  (2024-02-14T13:30:02Z)
• Pre-processing training data improves accuracy and generalisability of convolutional neural network based landscape semantic segmentation [2.8747398859585376]
  We trialled different methods of data preparation for CNN training and semantic segmentation of land use land cover (LULC) features within aerial photography over the Wet Tropics and Atherton Tablelands, Queensland, Australia. This was conducted through trialling and ranking various training patch selection sampling strategies, patch and batch sizes and data augmentations and scaling. We fully trained five models on the 2018 training image and applied the model to the 2015 test image with the output LULC classifications achieving an average of 0.84 user accuracy of 0.81 and producer accuracy of 0.87.
  arXiv  Detail & Related papers  (2023-04-28T04:38:45Z)
• Learning Where to Learn in Cross-View Self-Supervised Learning [54.14989750044489]
  Self-supervised learning (SSL) has made enormous progress and largely narrowed the gap with supervised ones. Current methods simply adopt uniform aggregation of pixels for embedding. We present a new approach, Learning Where to Learn (LEWEL), to adaptively aggregate spatial information of features.
  arXiv  Detail & Related papers  (2022-03-28T17:02:42Z)
• Data Augmentation with norm-VAE for Unsupervised Domain Adaptation [26.889303784575805]
  We learn a unified classifier for both domains within a high-dimensional homogeneous feature space without explicit domain adaptation. We employ the effective Selective Pseudo-Labelling (SPL) techniques to take advantage of the unlabelled samples in the target domain. We propose a novel generative model norm-VAE to generate synthetic features for the target domain as a data augmentation strategy.
  arXiv  Detail & Related papers  (2020-12-01T21:41:08Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.