search

tusen-ai / simpledet

A Simple and Versatile Framework for Object Detection and Instance Recognition
Apache License 2.0
3.08k stars 488 forks source link

How can we choose epoch? #311

Open vhnamgau opened 4 years ago

vhnamgau commented 4 years ago

It is difficult to choose the suite epoch for stopping training. Could you please give us some document regarding this task. Thanks

xchani commented 4 years ago

You may use TensorBoard to monitor the loss while training, and stop as it going steady. Here is the link of setting up: TensorBoard.

vhnamgau commented 4 years ago

Thanks Chenxia Han for your answer. TenserBoard is a tool to monitor : RcnnL1, RcnnAcc, RpnL1,RpnAcc. But how can I choose the epoch to stop training. We also need more document regarding RcnnL1, RcnnAcc, RpnL1,RpnAcc.

huangzehao commented 4 years ago

Hi, you can simply choose the last epoch. Since it is hard for network to overfit a large dataset like COCO. Training longer and you will get better performance.