Haoen Feng

Results 9 comments of Haoen Feng

请问如果用其后的卷积层吸收bias的话,因为没有BN层,所以是用来更新conv里的bias吗,但是conv的bias的维度与前一层BN的bias维度不一致的,应该怎么解决这个问题呢?

为什么不加入下一层(含BN)的bias,只加入到running_mean呢?

> > 为什么不加入下一层(含BN)的bias,只加入到running_mean呢? > > BN层没有bias BN层有两个参数,一个gamma,一个是shift,我说的bias是指shift的参数,感觉应该跟running_mean一起更新才对

> > 为什么不加入下一层(含BN)的bias,只加入到running_mean呢? > > BN层没有bias 谢谢分析,理清了我的思路。关于第四点里的mean,它是指batch_mean还是running_mean呢?如果是running_mean,按照官方文档的设置它是用于evaluation代替training时的shift的,training时并不参与,这样子training时怎么影响下一层的数据呢?谢谢

> > > > 为什么不加入下一层(含BN)的bias,只加入到running_mean呢? > > > > > > > > > BN层没有bias > > > > > > 谢谢分析,理清了我的思路。关于第四点里的mean,它是指batch_mean还是running_mean呢?如果是running_mean,按照官方文档的设置它是用于evaluation代替training时的shift的,training时并不参与,这样子training时怎么影响下一层的数据呢?谢谢 > > rolling_mean, 刚才说的操作是在训练完毕后的剪枝过程里,剪枝的时候不涉及训练 明白了,感激不尽!!!

@IMABUNNEH Have you solved your problem? I'm still getting nan even that I've set my warmup batches to 3 and reset the anchors. I used the pre-trained weight "full_yolo_backen.h5"

@letilessa Have you fix your problem? I always got Nan at the beginning of training even I changed the anchors and used the pre-trained weight. I used YOLOv2 as backend....

+1. Any suggestions will be appreciated.