Table 2.
Performance of Object Detection Method on Images Generated From Demographic Misspecification.
| Misspecification | bbox detection rate | Average width | Average number of bboxes | Precision | Recall | AUC |
|---|---|---|---|---|---|---|
| None (baseline) | 0.950 | 10.830 (var = 0.615, n = 1,978) |
1.027 (var = 0.063, n = 2,000) |
0.886 | 0.897 | 0.871 |
| m = 0.1 | 0.767 | 10.771 (var = 0.850, n = 774) |
0.787 (var = 0.198, n = 1,000) |
0.942 | 0.734 | 0.743 |
| m = 0.25 | 0.885 | 10.838 (var = 0.619, n = 906) |
0.953 (var = 0.181, n = 1,000) |
0.912 | 0.851 | 0.860 |
| m = 0.75 | 0.846 | 10.795 (var = 0.683, n = 876) |
0.881 (var = 0.115, n = 1,000) |
0.875 | 0.765 | 0.731 |
| m = 0.9a | 0.082 | 10.763 (var = 0.277, n = 213) |
0.213 (var = 0.168, n = 1,000) |
0.342 | 0.073 | 0.044 |
| gen = 25 | 0.874 | 10.824 (var = 0.624, n = 959) |
0.995 (var = 0.087, n = 1,000) |
0.814 | 0.799 | 0.771 |
| gen = 100 | 0.977 | 10.777 (var = 0.837, n = 996) |
1.013 (var = 0.025, n = 1,000) |
0.914 | 0.918 | 0.884 |
| Fst = 0a | 0.046 | 10.879 (var = 0.287, n = 717) |
1.262 (var = 1.173, n = 1,000) |
0.054 | 0.057 | 0.015 |
| Bottleneck (50%) | 0.953 | 10.858 (var = 0.498, n = 995) |
1.046 (var = 0.092, n = 1,000) |
0.872 | 0.895 | 0.860 |
| Bottleneck (10%) | 0.939 | 10.846 (var = 0.544, n = 990) |
1.021 (var = 0.047, n = 1,000) |
0.860 | 0.870 | 0.836 |
| Expansion | 0.945 | 10.809 (var = 0.700, n = 981) |
1.017 (var = 0.063, n = 1,000) |
0.887 | 0.889 | 0.865 |
| Contraction | 0.944 | 10.881 (var = 0.403, n = 987) |
1.042 (var = 0.088, n = 1,000) |
0.864 | 0.883 | 0.852 |
Further details of models presented in Materials and Methods, supplementary figure S1, Supplementary Material online. The two scenarios that perform poorly are marked (a).