Detect Early, Escalate Rarely: Anytime Detection of AI-Generated Video from the Compressed Bitstream
Quick Answer
This paper presents a novel approach for detecting AI-generated videos directly from compressed bitstreams, achieving an AUC of 0.64 with significantly lower compute costs—up to $7 imes$ less than traditional pixel CNNs.
Quick Take
The method allows for real-time detection with a monotone scoring system, improving accuracy from 0.75 to 0.78 by deferring 15% of clips.
Key Points
- Achieves AUC of 0.64 on GenVidBench with reduced compute requirements.
- Real-time detection uses motion field data from the codec instead of pixel decoding.
- Deferring 15% of clips increases accuracy from 0.75 to 0.78.
- Monotone scoring allows for anytime-valid thresholds in detection.
- No new detector introduced; focus is on reframing the detection task.
DeepSignal Analysis
What happened
A new method for detecting AI-generated videos from compressed bitstreams has been proposed, achieving an AUC of 0.64 while requiring significantly less computational power compared to traditional methods. This approach allows for real-time detection and improves accuracy by deferring some clips.
Key evidence
- The proposed method achieves an area under the curve (AUC) of 0.64 while using five orders of magnitude less compute than pixel CNNs on a CPU.
- By deferring 15% of video clips, the accuracy of detection improves from 0.75 to 0.78, with a significant reduction in compute costs.
- The method utilizes a monotone scoring system that allows for anytime-valid decision thresholds based on the data-dependent conditions.
Why it matters
This advancement in video detection technology could enhance the ability to identify AI-generated content in real-time, which is increasingly important as such content becomes more prevalent. The reduced computational requirements may also make this technology more accessible for various applications, including content moderation and verification.
What to watch
Paper Resources
📖 Reader Mode
~2 min readAbstract:Detectors for AI-generated video are evaluated offline. A clip is decoded to pixels and scored once, increasingly by a large vision-language model. Detection, however, is deployed online. We recast the task as streaming perception and score the motion field the codec already wrote into the bitstream. Reading that field is a parse, not a pixel-domain forward pass. Because the running aggregate is monotone, one end-calibrated threshold is anytime-valid at the data-dependent decision time. Recalibrating at each prefix is not. Escalation is priced in closed form. A compute budget maps to a deferral window, on a frontier monotone exactly where the deferral condition holds. On matched GenVidBench the codec stage reaches full-length AUC 0.64 at five orders of magnitude less compute than a pixel CNN, on CPU. Its gate holds the stopping-time false-positive rate at target while the real data match its calibration, and drifts above it under distribution shift. Deferring 15% of clips lifts accuracy from 0.75 to 0.78 at $7\times$ less compute (paired: McNemar $p<10^{-6}$). The stage-1 ordering replicates on AIGVDBench. We introduce no new detector. The contribution is the reframing, two guarantees, and the measured frontiers. Code, configurations, and evaluation splits: this https URL.
| Subjects: | Computer Vision and Pattern Recognition (cs.CV) |
| Cite as: | arXiv:2607.19476 [cs.CV] |
| (or arXiv:2607.19476v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2607.19476 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Mert Cakiroglu [view email]
[v1]
Tue, 21 Jul 2026 18:00:51 UTC (665 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.