对参考代码做了修改，将代码结构化，并使其可以正常训练，训练代码中没有进行测试。
Baidu's Densebox implemention with PyTorch used for multi-task learning of object detection and landmark(key-point) localization
extract code: ir6n
图片名字_label_[leftup_x][leftup_y][rightdown_x]_[rightdown_y].jpg
eg: det_2018_09_13_000002_label_54_105_171_153.jpg
训练图片均放在了dataset文件夹下面
python train.py
python test.py