How to test the model?

[ZhengLeon]

@DoubtedSteam
Hi~
How can I eval the model and get the result in your paper after training?

Thanks!

[DoubtedSteam]

set epoch to 0 and resume the checkpoint
you can add the resume: path/to/the/checkpoint into yaml file

[ZhengLeon]

Thanks a lot!
If I enable the resume in training process, how can I update the begining epoch num using the resume one instead of epoch0?

[DoubtedSteam]

edit num_epoch in yaml file

[ZhengLeon]

Just edit the "num_epoch" in SYSU.yml doesn't work, isn't this "num_epoch" represent the max epoch?
(I wanna change the first training epoch num to the resume epoch num so the old checkpoints could avoid being covered.)

[DoubtedSteam]

It is for max epoch

And it is really strange, When training finishes, it will automatically testing.

[ZhengLeon]

In Chinese:
恢复训练时的epoch是从0开始的，这样似乎会覆盖之前的checkpoint,如何修改使重新训练的epoch是从resume的轮数开始呢？

[DoubtedSteam]

I didn't achieve this function.
You can add a start_epoch parameter for def get_trainer() in engine/init.py

实际上太麻烦了，因为学习率的step要一次次调用，所以没写这个功能

[ZhengLeon]

strategy_cfg.num_epoch = 60
num_epoch: 140
Dose only the second one make sense?
嗯嗯 如果训练过程中断了epoch的数量就只能从0开始比较方便吗