Will you or did you try Group Normalization

[XCRobert]

According to your paper, BN plays a critical role training from scratch. Did you try Group Normalization ? Does Group Normalization have better perfomance than some pretrained ones. According your assumption,it may also works.

[KimSoybean]

@XCRobert Hello, I have tried GN on ScratchDet on mmdetection and get 1 mAP lower result than BN. Sorry I haven't found the reason but I think it is relevant to learning rate. But more details should be given in analysis with gradients. I will do that after I get my IELTS testing passed. Thank you!