Deepfake detection often falters on edge devices where compute budgets are slim and video quality is uneven. To address this mismatch, AdaGate-DF routes frames through a dual multi-exit architecture keyed to image-quality cues, allowing clearer samples to terminate early and cut latency. Across benchmark evaluations on Celeb-DF, the framework records an AUC of 0.9370 with low inference overhead, climbing to 0.9708 at 384×384 resolution. By replacing fixed inference pipelines with adaptive routing, it provides a practical path for authenticating visual media under real-world bandwidth and resource constraints.
No heat snapshots are available in the last 24 hours.