SunAnLan

这个是为了防止出现全学偏了，我觉得加了没关系，反正也不会回传给detnet的参数。

1. 150 can help converge. if you use 127.5, the tanh must be close to 1 to output a 255, where the derivative is pretty low. 2. that is quite...

@xuhuaren could you please tell me the version of your pytorch and cuda? As I just tried with the config like yours, there comes an error `RuntimeError: Exporting the operator...

@dat-ai It'Okay ~ I also ran into a problem.