对nanodet算法进行模型剪枝
对模型进行剪枝，在自定义数据集上，剪枝前后的指标为为：
剪枝前
Params: 9.31 M , MACs : 11.44 G
| class | AP50 | mAP | class | AP50 | mAP |
|:----------|:-------|:------|:---------|:-------|:------|
| car | 55.9 | 40.8 | bus | 32.4 | 19.4 |
| truck | 66.5 | 46.4 | person | 32.4 | 20.2 |
| bicycle | 0.0 | 0.0 | tricycle | 0.0 | 0.0 |
| motorbike | 34.3 | 15.1 | | | |
'mAP': 0.20263765407418965,
'AP_50': 0.31637914695810804,
'AP_75': 0.21580095003729519,
'AP_small': 0.02556794488193019,
'AP_m': 0.26315913744876807,
'AP_l': 0.6280771993427688剪枝后，微调 50 个epoch
after speed Params: 6.22 M , MACs 7.51 G
| class | AP50 | mAP | class | AP50 | mAP |
|:----------|:-------|:------|:---------|:-------|:------|
| car | 51.9 | 35.2 | bus | 0.0 | 0.0 |
| truck | 63.4 | 42.3 | person | 31.0 | 20.2 |
| bicycle | 0.0 | 0.0 | tricycle | 0.0 | 0.0 |
| motorbike | 29.3 | 14.8 | | | |
mAP: 0.16087444613389984
AP_50: 0.2509137585063215
AP_75: 0.17434789336310114
AP_small: 0.009993173108280119
AP_m: 0.20596705780074465
AP_l: 0.5062155473672427剪枝前为全部数据进行训练，剪枝后微调的数据是不包括全部标签、较小的数据集