Image-level manipulation detection on Columbia
98.4AUCMVSS-Net++
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| MVSS-Net++Decision threshold=0.52021.12 | 98.4 | 96.7 | 89.6 | 93 | — | |
| MVSS-NetDecision threshold=0.52021.04 | 98 | 66.9 | 100 | 80.2 | — | |
| MVSS-NetDecision threshold=0.52021.12 | 98 | 66.9 | 100 | 80.2 | — | |
| CAT-NetDecision threshold=0.52021.12 | 97.1 | 87.2 | 96.2 | 91.5 | — | |
| TruFor2026.05 | 92.4 | — | — | — | 89.1 | |
| CAT-Net2026.05 | 91.6 | — | — | — | 88.2 | |
| FRAMESelection=Learned, Fusion=Learned2026.05 | 90.8 | — | — | — | 86.7 | |
| ManTraNet2026.05 | 89.7 | — | — | — | 85.8 | |
| XGB-Ensemble (pyIFD)Fusion=XGB-Ensemble2026.05 | 87.6 | — | — | — | 83.4 | |
| RF-Ensemble (pyIFD)Fusion=RF-Ensemble2026.05 | 86.8 | — | — | — | 82.6 | |
| MMFusion2026.05 | 86.3 | — | — | — | 82.1 | |
| Heuristic-K + uniformSelection=Heuristic, Fusion=Uniform2026.05 | 85.6 | — | — | — | 81.3 | |
| Best single pyIFDSelection=Best single2026.05 | 84.1 | — | — | — | 79.8 | |
| Random-K + uniformSelection=Random, Fusion=Uniform2026.05 | 82.8 | — | — | — | 78.2 | |
| Uniform-all pyIFDSelection=None (all), Fusion=Uniform2026.05 | 81.7 | — | — | — | 77.4 | |
| CR-CNNDecision threshold=0.52021.04 | 78.3 | 96.1 | 24.6 | 39.2 | — | |
| CR-CNNDecision threshold=0.52021.12 | 78.3 | 96.1 | 24.6 | 39.2 | — | |
| FCNDecision threshold=0.52021.04 | 76.2 | 95 | 32.2 | 48.1 | — | |
| FCNDecision threshold=0.52021.12 | 76.2 | 95 | 32.2 | 48.1 | — | |
| ManrTra-NetDecision threshold=0.52021.04 | 70.1 | 100 | 0 | 0 | — | |
| ManrTra-NetDecision threshold=0.52021.12 | 70.1 | 100 | 0 | 0 | — | |
| H-LSTMDecision threshold=0.52021.12 | 50.6 | 100 | 1.1 | 2.2 | — | |
| GSR-NetDecision threshold=0.52021.04 | 50.2 | 100 | 1.1 | 2.2 | — | |
| GSR-NetDecision threshold=0.52021.12 | 50.2 | 100 | 1.1 | 2.2 | — | |
| SPANDecision threshold=0.52021.12 | 50 | 100 | 0 | 0 | — |