Table 6 Effect of modeling choices (Tanzania data, coarse supervision).
From: Fine-grained population mapping from coarse census counts and open geodata
Method | LogL1 | Occ. | Aug. | \(R^2\) \(\uparrow\) [%] | MAE \(\downarrow\) | MAPE \(\downarrow\) [%] |
|---|---|---|---|---|---|---|
Baseline | 71.4 ± 0.3 | 4130 ± 80 | 27.5 ± 0.9 | |||
+ Log \(L_1\) loss | \(\checkmark\) | 73.2 ± 0.3 | 3580 ± 30 | 23.0 ± 0.2 | ||
+ Occupancy rate | \(\checkmark\) | 81.4 ± 1.2 | 3820 ± 180 | 26.8 ± 1.2 | ||
+ Augmentation | \(\checkmark\) | 73.0 ± 1.2 | 3820 ± 30 | 25.0 ± 0.4 | ||
Pomelo | \(\checkmark\) | \(\checkmark\) | \(\checkmark\) | 85.8 ± 0.7 | 3100 ± 30 | 21.7 ± 0.3 |
