Table 1 Effect of super net node N on precision and weight
Algorithms | Number of nodes N | AP50/% | Recall/% | F1/% | Weight size/MB | Single image detection time/ms |
|---|---|---|---|---|---|---|
Faster-RCNN +FPN | – | 84.9 | 91.7 | 88.2 | 322.54 | 161 |
Ours (Faster-RCNN + NAS-FPN) | 2 | 83.7 | 91.2 | 87.3 | 265.66 | 135 |
3 | 85.3 | 91.7 | 88.4 | 266.69 | 140 | |
4 | 87.3 | 91.2 | 89.2 | 270.65 | 140 | |
5 | 88.3 | 91.2 | 91.0 | 277.18 | 147 | |
6 | 88.0 | 91.3 | 90.8 | 304.19 | 158 | |
7 | 86.7 | 93.6 | 90.4 | 311.17 | 166 |