SmartCar(SSD-mobilenetv2)
- this repo is forked from https://github.com/amdegroot/ssd.pytorch. Implemented by pytorch.
Contributions:
- 基于SSD大框架加入mobilenetv2以实现在嵌入式GPU开发板上的视频流识别
- 增加 focal loss.
- 增加detection.py来单纯验证SSD-mobilenetv2算法
- 提供rs.py来验证D415传感器性能
- 在detection_D415.py中,加入D415双目摄像头来实现深度预估,以实现物体的动态抓取
- 提供 visdom可视化,使用python -m visdom.server即可观察收敛
result(train on voc 2007trainval + 2012, test on voc 2007test):
- ssd-mobielnetv2 (this repo): 70.27%. (without focal loss).
- ssd-mobielentv1: 68.% (without COCO pretaining), 72.7% (with COCO pretraining) https://github.com/chuanqi305/MobileNet-SSD.
- ssd-vgg16 (paper): 77.20%.
pretrained model and trained model: