search

xuexingyu24 / License_Plate_Detection_Pytorch

A two stage lightweight and high performance license plate recognition in MTCNN and LPRNet
Other
641 stars 171 forks source link

train_accuracy #4

Open nightfuryyy opened 5 years ago

nightfuryyy commented 5 years ago

i tried to train my model on my plate data.

"loss: 0.0000, train_accuracy: 0.1875, time: 0.28 s/iter, learning rate: 0.0001"

loss so small, but train_accuracy always <= 0.4 . I trained 5hour and can't get better result. How can i fix that ?

xuexingyu24 commented 5 years ago

@nightfuryyy how did you initialize the weights during training ? I will recommend you to use pretrained weights from sirius's repo as an initializer.

nightfuryyy commented 5 years ago

my classnum is 31. So i can't use his pretrained weights. Btw i generated gray images for training.

wait1988 commented 5 years ago

只是用了ccpd来训练吗，从模型测试效果来看，感觉数据很不均衡 @xuexingyu24

kunyao2015 commented 4 years ago

same problem , initialize the weights random

persuelx commented 4 years ago

@nightfuryyy @kunyao2015 your problem has solved?I had the same problem when I trained my custom data。