红绿灯分类

环境：mxnet, gluon, mac,基于ImageNet预训练模型transfer learning, finetunning

训练

python train.py 测试结果：

epoch 1, loss 5.7411, train acc 0.468, test acc 0.651, time 123.7 sec
epoch 2, loss 4.6547, train acc 0.517, test acc 0.860, time 119.4 sec
epoch 3, loss 0.7964, train acc 0.769, test acc 0.884, time 123.1 sec
epoch 4, loss 0.9767, train acc 0.725, test acc 0.872, time 142.4 sec
epoch 5, loss 0.5056, train acc 0.806, test acc 0.802, time 120.0 sec
epoch 6, loss 0.8685, train acc 0.711, test acc 0.884, time 126.5 sec

推理

python inference.py 测试结果：classify result: red 数据太少，精确度不是很高

参考

http://zh.gluon.ai/chapter_computer-vision/fine-tuning.html?highlight=hotdog

https://gluon-cv.mxnet.io/build/examples_classification/demo_cifar10.html

预模型模型

链接:https://pan.baidu.com/s/1Pp5Vf5oBiWzsjP08Op36LQ 密码:8br1