The experiment was trained on 40,000 data sets, and 40,000 iterations were performed. The test MAP used 15,360 images
| Model | MAP |
|---|---|
| YOLO-V3-tiny | 42% |
| YOLO-V3+GIOU | 62.5% |
| YOLO-V3+Soft-NMS | 65.7% |
| FCOS-ResNet50 | 65.38%(5 epoch) |
| FCOS-mobilenet v2 | 75% (90000 iters) |
