Oriel — AI & technology worth noticing, every day.

Beyond the Last Layer: Global Inter-Layer Aggregation for AI-Generated Image Detection

Paper · https://arxiv.org/abs/2611.00017

arXiv:2611.00017v1 Announce Type: new Abstract: The rapid development of generative models has made AI-generated images increasingly close to real images. It has also brought several challenges to detectors, including rapid generator updates, large changes in data distribution, and limited cross-dataset generalization. Many existing methods based on pretrained vision encoders primarily use the global representation from the last layer, potentially leaving useful information from intermediate layers underexplored. To address this issue, we propose Global Inter-Layer Aggregation (GILA). GILA extracts global representations from different depths of a frozen encoder, models interactions among layer representations through inter-layer self-attention, and adaptively aggregates the resulting contextualized representations. The method is not tied to a single pretraining objective and does not require fine-tuning the pretrained encoder. We evaluated GILA on three vision backbones with different pretraining objectives, ten seen generators, and three external datasets. The experimental results show that GILA makes better use of multi-layer information in pretrained encoders and improves detection performance and cross-dataset generalization.

First seen by Oriel on .

AI summary

The paper introduces Global Inter-Layer Aggregation (GILA), which combines representations from multiple depths of a frozen vision encoder to improve AI-generated image detection and cross-dataset generalization.

  • GILA extracts global representations at different encoder depths, relates them with inter-layer self-attention, and adaptively aggregates the contextualized representations.
  • The method does not require fine-tuning the pretrained encoder and is not limited to one pretraining objective.
  • The evaluation covers three vision backbones, ten seen generators, and three external datasets.
  • The reported results show improved detection performance and cross-dataset generalization, though the abstract provides no quantitative metrics.

#image detection #computer vision #self-attention #foundation models

Generated by gpt-5.6-sol. Check the original for details.

Aliases

SourceExternal IDURL
arXiv AI Papersoai:arXiv.org:2611.00017v1https://arxiv.org/abs/2611.00017 Read discussion