基于多方位感知深度融合检测头的目标检测算法
|
|
包晓安,彭书友,张娜,涂小妹,张庆琪,吴彪
|
Object detection algorithm based on multi-azimuth perception deep fusion detection head
|
|
Xiao’an BAO,Shuyou PENG,Na ZHANG,Xiaomei TU,Qingqi ZHANG,Biao WU
|
|
| 表 1 不同目标检测器在MS-COCO2017数据集上使用MdfHead的结果 |
| Tab.1 Results of applying MdfHead to different object detectors on MS-COCO2017 dataset |
|
| 检测器 | 方法 | Np/106 | FLOPs/109 | AP/% | AP50/% | AP75/% | FPS/(帧·s−1) | | RetinaNet(R101) | Baseline | 56.961 | 282.91 | 38.5 | 57.6 | 41.0 | 23.2 | | RetinaNet(R101) | MdfHead | 58.013 | 283.42 | 41.4 | 60.7 | 43.6 | 22.9 | | FCOS(R101) | Baseline | 51.287 | 248.26 | 39.1 | 58.3 | 42.1 | 23.1 | | FCOS(R101) | MdfHead | 52.117 | 249.19 | 41.7 | 61.0 | 44.1 | 23.0 | | CenterNet(R50) | Baseline | 32.293 | 179.99 | 40.2 | 58.3 | 43.9 | 32.9 | | CenterNet(R50) | MdfHead | 33.311 | 183.37 | 42.0 | 60.4 | 45.8 | 32.5 | | ATSS(R101) | Baseline | 51.283 | 252.52 | 41.5 | 59.9 | 45.2 | 23.4 | | ATSS(R101) | MdfHead | 52.108 | 253.04 | 44.9 | 63.6 | 49.1 | 23.2 | | PAA(R101) | Baseline | 51.435 | 255.11 | 42.6 | 60.8 | 46.6 | 20.5 | | PAA(R101) | MdfHead | 52.303 | 255.84 | 45.2 | 63.2 | 48.4 | 20.3 |
|
|
|