Skip to content
This repository has been archived by the owner on Apr 6, 2024. It is now read-only.

too many anchor boxes #7

Closed
ruikun opened this issue Jun 7, 2018 · 5 comments
Closed

too many anchor boxes #7

ruikun opened this issue Jun 7, 2018 · 5 comments

Comments

@ruikun
Copy link

ruikun commented Jun 7, 2018

大神你好,
我加载了你提供的VOC数据集的weight继续训练我自己的数据集(只有一个类:house)。
进行到3000+ steps的时候loss突然变成nan,此前在1.0上下波动。
我加载了最新的ckpt,test结果图片上遍布密集的anchor box,如图:
image
想请教大佬一下大概是哪里有问题,请问您提供的weight也是通过这个项目训练出来的吗?

@ruikun ruikun changed the title 检测框数量 too many anchor boxes Jun 7, 2018
@ruikun ruikun mentioned this issue Jun 8, 2018
@Nine9Nine
Copy link

可能是训练好的都是20类的吧

@leeyoshinari
Copy link
Owner

别恢复所有层的权重,调整一下学习速率,重新训练试一下

@ruikun
Copy link
Author

ruikun commented Jun 10, 2018

我只有最后一层的weights 变量没有加载,毕竟我自己的数据集只有一个类嘛。
速率的话,已经是0.00001了

@leeyoshinari
Copy link
Owner

最后几层的权重都不要恢复,你再试一下。

@ruikun
Copy link
Author

ruikun commented Jun 10, 2018

请问你是从scratch开始用voc训练的吗?一开始梯度爆炸的问题是怎么解决的?

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants