The results of running demo.py create too many boxes.

[DaehanKim-Korea]

I replaced the network among the codes in the Repo. (resnet101) Losses are decreasing normally. The results of running demo.py create too many boxes. Is it too early to judge / now iteration is 6000

[Linda-L]

why can't I find demo.py?

[linyanhly]

I think you should modify the threshold of the output,then you can see the lagerst Confidence level boxes in the output images

[Qiyuan-Z]

@koreadhkim I have the same problem. Have you solved it？

[Ckkmissing]

It seems that I found something. I used the Adam optimizer, and the above results often appeared, but changed to SGD and it didn’t appear again. Of course,the model maybe is too large, maybe you can change it to a smaller one.

[zhangnan-hust]

I have the same problem. I trained my ssd netwok by VOC2007_trainval. After 200 epochs, I test the ssd with 0.05 threshold and got too many boxes. These boxes include many wrong classes.

[software93]

I have same problem please dose anyone of you solved this problem?