Frame-level Video Manipulation Detection on Merged authentic and manipulated frames (evaluation)
83.12Global AccuracyDINOv3 Feature Similarity framework
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| DINOv3 Feature Similarity frameworkBackbone=DINOv3, Protocol=Unsupervised, Approach=Temporal cosine distance between consecutive frames2026.06 | 83.12 | 83 | 53.64 | 65.16 |