Mang Ning

this pre-trained feature gonna save lives...

> 我也是 prob_loss 几乎为0 兄弟，你的 giou_loss 如何呢？最终降到了多少？

> ``` > def call(self, x, training=False): > if not training: > training = tf.constant(False) > training = tf.logical_and(training, self.trainable) > return super().call(x, training) > ``` > > 代码training =...

> Any update on this? if your giou firstly turned out nan, it is likely that there is something wrong in the defined giou function. In my experiment, I found...

同问，训练自己的模型的时候，backbone 部分的权重为啥不直接加载官方已经训练好的权重呢

> 可以的，但需要在加载模型的时候注释掉原有的模型加载方式，使用如45行的方式加载 > […](#) > ------------------ 原始邮件 ------------------ 发件人: "forever208"

感谢 llllala，另外想再问一下，加载权重后准备训练，我想在训练阶段 freeze bn 层，请问需要咋操作呢。bn 的子类我看得不太明白，还请赐教。感恩~！

following..

1. the reason you got this issue is that you miss some txt labels file in the path `darknet/data/obj` For example, you included 0001.txt in `darknet/data/train.txt`, but 0001.txt doesn't' exist...

The problem of NaNs still exists with this change. > > > Perhaps this bug is related to the issue here: #44 > > If so, perhaps we could try...