Research map / Video / Video segmentation and tracking
SPLIT: Training-Free AI-Generated and Partially Edited Video Detection via Spatial Patch‑Level Incoherence and Temporal Roughness
Jongyeop Hyun, Hyounghun Kim
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use patch-level tokens instead of global CLS embeddings for TTR (ablation) | TTR · FakeParts Overall | Fake Recall at FPR 0.1% (%): 54.21 -> 68.29 | Improved |
| add LSMI to patch-level TTR (ablation) | SPLIT · FakeParts Overall | Fake Recall at FPR 0.1% (%): +6.16 | Improved |
| increase gamma from 1 to 8 in the fusion rule (ablation) | SPLIT · FakeParts Overall | Fake Recall at FPR 0.1% (%) | Improved |
| increase gamma from 8 to 16 (ablation) | SPLIT · FakeParts Overall | Fake Recall at FPR 0.1% (%) | No clear effect |
| use SPLIT patch-level scoring instead of D3 with same XCLIP-B/16 encoder | D3 · FakeParts Overall | Fake Recall at FPR 0.1% (%): 2.28 -> 74.45 | Improved |
| use XCLIP-B/16 instead of XCLIP-B/32 encoder (ablation) | SPLIT · FakeParts Overall | Fake Recall at FPR 0.1% (%): 60.23 -> 74.45 | Improved |
| use CLIP-B/16 instead of CLIP-B/32 encoder (ablation) | SPLIT · FakeParts Overall | Fake Recall at FPR 0.1% (%): 47.37 -> 71.47 | Improved |
| apply Gaussian blur sigma=2 to input videos | SPLIT · FakeParts Overall | Fake Recall at FPR 0.1% (%): 74.45 -> 55.42 | Hurt |
| apply JPEG compression q=80 | SPLIT · FakeParts Overall | Fake Recall at FPR 0.1% (%): 74.45 -> 64.40 | Hurt |
| apply vertical flip to input videos | SPLIT · FakeParts Overall | Fake Recall at FPR 0.1% (%): 74.45 -> 66.92 | Hurt |
| apply horizontal flip to input videos | SPLIT · FakeParts Overall | Fake Recall at FPR 0.1% (%): 74.45 -> 75.73 | No clear effect |
| replace D3 with SPLIT on ResNet18 encoder | D3 · FakeParts Overall | Fake Recall at FPR 0.1% (%): 1.93 -> 61.23 | Improved |
Closest papers by meaning
- AI-Generated Video Detection via Perceptual Straightening (NeurIPS 2025 · similarity 0.86)
- MixForensics: Blend Before You Encode for Generalizable AI-Generated Video Detection (NeurIPS 2026 · similarity 0.86)
- Physics-Driven Spatiotemporal Modeling for AI-Generated Video Detection (NeurIPS 2025 · similarity 0.86)
- Preserving Forgery Artifacts: AI-Generated Video Detection at Native Scale (ICLR 2026 · similarity 0.86)
- Rethinking the Readout: Unlocking Video Backbones for AI-Generated Video Detection (NeurIPS 2026 · similarity 0.85)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
