Table 1.
Comparison of performance of architecture on test set
| Feature extractor | Object detector | mAP | AR | F1 Score |
|---|---|---|---|---|
| Human eye (embryologists) | Occipital cortex | 0.925a | 0.642 | 0.758 |
| MobileNet V2 (CNN) | SSD | 0.741 | 0.376 | 0.499 |
Reported mAP is at 0.5 IOU, and AR is at 100 detections per image. aWe report the raw precision here. Under the assumption that humans predict with 100% confidence, raw precision is equivalent to mAP. mAP: mean average precision; AR: average recall; SSD: single-shot detector; IOU: intersection-over-union ratio