The configuration of parameters can't find the rule

[elfpattern]

When I see the code, I can't figure the fule of parameters, Can you explain??

[liyunsheng13]

s, n, c, ks represents the stride, number of repeated layers, network width and kernel sizes. The multiplication of c1 and c2 is the expansion between input channels to the hidden dimension. g1 and g2, g3 and g4 are basically the same parameters that represent the group number used in the macro-factorized convolution. The rest three parameters are used to decide whether the dynamic shift-max is used or not, while the last parameter is the hyperparameter contained in dynamic shift-max to adjust the reduction ratio.

[elfpattern]

@liyunsheng13 Thx for replying, I have another question, I convert the pytorch model into mnn or onnx model, The speed has no become more faster, why?

[thgpddl]

你好，我在阅读源代码时，发现按照上图cfgs总共有12层，但是论文Table1中M3模型是1stem+11MicroBlock+1claaify，我想请问为什么论文中11MicroBlock和cfgs中的12层对应不上呢？