about some details

[foralliance]

@KimSoybean HI
有几个疑问：

1. 在分析BN作用时，是基于SSD+VGG16．能不能这么认为，只要是一个包含了BN层的网络，将其作为detector的底层网络时，该detector都可以train from scratch？

2. 4.1节指出：remove the L2 Normalization．是不是因为BN层的存在可以代替这个L2 Normalization？

[KimSoybean]

@foralliance Hi！I think we should speak English because your questions may help the people in other countries.

1. No. When you train the model with large input image size (e.g., 800x1300), the batch-size will reduce to 1-2 due to the limited GPU memory. Then the effect of BN will be constrained. If so, please replace BN with GN. GN does not care batch-size.

2. This question is very complex. I can just answer your question that they are not relevant.

[foralliance]

@KimSoybean
many many thanks!

[dby2017]

Have you tried using Root-ResNet-18/34 on faster-rcnn？

[KimSoybean]

@dby2017 Hello, I haven't tried Root-Res on faster r-cnn. Maybe it is not very effective due to the large input resolution (small objects will be larger than before).