Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

EA-Swin: An Embedding-Agnostic Swin Transformer for AI-Generated Video Detection

Hung Mai, Loi Dinh, Duc Hai Nguyen, Dat Do, Luong Doan, Khanh Nguyen Quoc, Huan Vu, Phong Ho, Naeem Ul Islam, Tuan Do

TL;DR

An Embedding-Agnostic Swin Transformer that models spatiotemporal dependencies directly on pretrained video embeddings via a factorized windowed attention design, making it compatible with generic ViT-style patch-based encoders and establishing a scalable and robust solution for modern AI-generated video detection.

Abstract

Recent advances in foundation video generators such as Sora2, Veo3, and other commercial systems have produced highly realistic synthetic videos, exposing the limitations of existing detection methods that rely on shallow embedding trajectories, image-based adaptation, or computationally heavy MLLMs. We propose EA-Swin, an Embedding-Agnostic Swin Transformer that models spatiotemporal dependencies directly on pretrained video embeddings via a factorized windowed attention design, making it compatible with generic ViT-style patch-based encoders. Alongside the model, we construct the EA-Video dataset, a benchmark dataset comprising 130K videos that integrates newly collected samples with curated existing datasets, covering diverse commercial and open-source generators and including unseen-generator splits for rigorous cross-distribution evaluation. Extensive experiments show that EA-Swin achieves 0.97-0.99 accuracy across major generators, outperforming prior SoTA methods (typically 0.8-0.9) by a margin of 5-20%, while maintaining strong generalization to unseen distributions, establishing a scalable and robust solution for modern AI-generated video detection.

EA-Swin: An Embedding-Agnostic Swin Transformer for AI-Generated Video Detection

TL;DR

An Embedding-Agnostic Swin Transformer that models spatiotemporal dependencies directly on pretrained video embeddings via a factorized windowed attention design, making it compatible with generic ViT-style patch-based encoders and establishing a scalable and robust solution for modern AI-generated video detection.

Abstract

Recent advances in foundation video generators such as Sora2, Veo3, and other commercial systems have produced highly realistic synthetic videos, exposing the limitations of existing detection methods that rely on shallow embedding trajectories, image-based adaptation, or computationally heavy MLLMs. We propose EA-Swin, an Embedding-Agnostic Swin Transformer that models spatiotemporal dependencies directly on pretrained video embeddings via a factorized windowed attention design, making it compatible with generic ViT-style patch-based encoders. Alongside the model, we construct the EA-Video dataset, a benchmark dataset comprising 130K videos that integrates newly collected samples with curated existing datasets, covering diverse commercial and open-source generators and including unseen-generator splits for rigorous cross-distribution evaluation. Extensive experiments show that EA-Swin achieves 0.97-0.99 accuracy across major generators, outperforming prior SoTA methods (typically 0.8-0.9) by a margin of 5-20%, while maintaining strong generalization to unseen distributions, establishing a scalable and robust solution for modern AI-generated video detection.
Paper Structure (27 sections, 3 equations, 10 figures, 13 tables)

This paper contains 27 sections, 3 equations, 10 figures, 13 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. AI-generated video detection.
  4. Benchmark datasets
  5. Method
  6. Video Embedding Representation
  7. Factorized Spatiotemporal Swin Transformer
  8. Temporal Swin Attention.
  9. Spatial Swin Attention.
  10. Block Composition.
  11. Token Aggregation and Classification
  12. EA-Video Dataset
  13. Dataset Curation
  14. Dataset Composition
  15. Experiment
  16. ...and 12 more sections

Figures (10)

  • Figure 1: Sampled video frames from some AI video generators. Top: Recent generators produce high-quality visuals and realistic motion, closely resembling real videos. Bottom: Earlier models show clear artifacts, distorted content, and unnatural motion.
  • Figure 2: Spatiotemporal window shifting mechanism. The input video is first partitioned into non-overlapping local windows (1) along spatial and temporal dimensions. To enable cross-window interaction and enhance global context modeling, the windows are then shifted (2) spatially across adjacent regions and temporally across neighboring frames.
  • Figure 3: Temporal Swin attention.
  • Figure 3: Ablation on vision backbone
  • Figure 4: Spatial Swin attention.
  • ...and 5 more figures