Qualitative analysis suggests that PatchHead reduces class-conditional domain discrepancy, and redirects the representation from content-dominated saliency toward spatially distributed authenticity evidence, which provides a representation-level account of why spatial patch aggregation transfers more reliably across generators and datasets than a single CLS-based global representation.
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.
Visual classifiers are expected to generalize under data shifts, target shifts, and their combinations, yet most existing methods focus on domain invariance while failing to address intra-image predictive sufficiency. We investigate the structural hypothesis that each image contains a sample-adaptive oracle intra-image predictive subset sufficient for label prediction, while the remaining patches form non-essential complementary context that may correlate with the label. The theoretical analysis shows that restricting prediction to this oracle subset preserves the Bayes risk achievable by the full-patch representation while admitting a complexity bound that tightens with the oracle-subset size. Based on this view, we propose PatchGen, a text-free module that learns a sample-dependent soft predictive-subset mask as a task-driven proxy for the unobserved oracle subset mask. Specifically, histopathology visualizations suggest that PatchGen assigns higher scores to tumor-consistent regions than to some frequently co-occurring inflammatory context. Extensive experiments on natural and histopathological image benchmarks spanning all three shift settings show that PatchGen improves average performance over matched-backbone baselines in most evaluated configurations, enhances generalization to unknown classes, and remains competitive with vision-language methods without text supervision.
A data-generation pipeline that captions real photographs with a vision–language model and regenerates them with modern text-to-image systems, producing semantically aligned real/synthetic pairs that isolate generative artifacts from image content is described.
Fine-grained visual classification relies on subtle local cues and is highly sensitive to input resolution, yet practical deployment often constrains image size and inference cost. Existing low-resolution recognition methods often depend on additional network structures or complex training modules, limiting deployment simplicity. This study evaluates a simple cross-resolution knowledge distillation (KD) strategy under a fixed low-compute constraint. The teacher receives higher-resolution inputs to provide richer class-discriminative visual knowledge, whereas the student is trained at 160×160 and is the only model retained for deployment. Both networks use ResNet-18, allowing resolution-aware supervision to be studied without changing the inference architecture. On the CUB-200-2011 dataset, the 160×160 baseline achieved 66.01±0.53% test accuracy. The 224→160 KD configuration improved accuracy to 70.15±0.63%, while the final 384→160 configuration with T = 4 , α = 0.75 , and a 30-epoch cosine schedule reached 71.60±0.68% accuracy and 71.56±0.62% macro-F1. The results also show that a stronger teacher does not automatically produce a stronger student: transfer quality depends on the balance between teacher supervision strength and student optimization. These findings support cross-resolution KD as a practical way to recover fine-grained information during training while preserving low-resolution inference cost.
Chen-Qiang Li· Applied and Computational En...· 0 citations
The proposed GlobalForge improves average BAcc on 8 in-the-wild benchmark groups by $\mathbf{5.89\%}$ over the previous state-of-the-art, and is clearly ahead of representative baselines on RealDeg-Bench under both single and compound degradations.
Manni Cui, Ruiqi Liu, Dianyuan Zou et al.· arXiv.org· 0 citations
Across cross-generator, post-processing, and in-the-wild benchmarks, PE-SPC surpasses the previous DINOv3 baseline and achieves new state-of-the-art results.
Wei-Han Cai, Hao Tan, Zichang Tan et al.· 0 citations
Vision models do not form a representation at once; each block revises it. We ask whether the resulting computation path contains evidence that the final representation discards, and whether that evidence improves OOD detection and image classification on clean and shifted data. Unlike approaches that treat intermediate layers as separate snapshots, we retain sample identity across depth and study the transformations connecting successive states. We separate class-coherent transport from input-specific innovation, and coordinate movement from relational reorganization. Across supervised, self-supervised, vision--language, hierarchical, and convolutional encoders, these paths show strong sample-specific continuity and architecture-specific depth profiles that recur across datasets. They are also practically useful. An ID-only transition-surprise score complements strong final-state detectors, reducing FPR95 in 131/152 non-saturated comparisons on a balanced OpenOOD grid; gains are largest for visually disruptive and semantically far shifts, and remain positive on near-OOD for most detectors. Frozen update probes improve 71/72 clean model--dataset cases, while shifted-data gains vary with architecture and corruption type. Computation paths therefore provide a broadly useful reliability signal whose value is determined jointly by model organization and the shift encountered.
Ignacio M. De la Jara, Cristian Rodriguez-Opazo, Hamed Damirchi et al.· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
