Global Source Attribution on MLAAD Closed-Set IID (test)
95.44F1-Macro (Seen IID)Strategy 1 (Orthonormal Prototypes)
Evaluation Results
| Method | Links | |
|---|---|---|
| Strategy 1 (Orthonormal Prototypes)MSE (λ)=1.02026.07 | 95.44 | |
| Strategy 2.A (Factorized Prototypes)MSE (λ)=1.02026.07 | 95.36 | |
| Strategy 1 (Orthonormal Prototypes)MSE (λ)=0.052026.07 | 95.34 | |
| Strategy 1 (Orthonormal Prototypes)MSE (λ)=0.12026.07 | 95.31 | |
| Strategy 2.A (Factorized Prototypes)MSE (λ)=0.12026.07 | 95.21 | |
| Strategy 1 (Orthonormal Prototypes)MSE (λ)=0.02026.07 | 95.2 | |
| Strategy 2.A (Factorized Prototypes)MSE (λ)=0.052026.07 | 95.17 | |
| Baseline (ArcFace)MSE (λ)=–2026.07 | 95.11 | |
| Strategy 2.B (Residual Modeling)MSE (λ)=1.02026.07 | 95.09 | |
| Strategy 2.B (Residual Modeling)MSE (λ)=0.12026.07 | 94.99 | |
| Strategy 2.A (Factorized Prototypes)MSE (λ)=0.02026.07 | 94.93 | |
| Strategy 2.B (Residual Modeling)MSE (λ)=0.052026.07 | 94.62 | |
| Strategy 2.B (Residual Modeling)MSE (λ)=0.02026.07 | 94.51 |