恩，我就问一下啊，我不是很懂手机和电脑不同之处，没别的意思
我电脑运行是 9.25ms 环境是 1080ti i7-7700k ubuntu

#我在MoGa_A.py的末尾加上这几句话，如果备注#model.load_state_dict那么是可以正常运行的，但是加载权重就报错了
if name == 'main':
from torchsummary import summary
model = MoGaA()
model.load_state_dict(torch.load('MoGaA.pth.tar'))#加载权重
summary(model,(3,224,224))
#RuntimeError: Error(s) in loading state_dict for MoGaA:

Missing key(s) in state_dict: "stem.0.weight", "stem.1.weight", "stem.1.bias",

What are the instructions for running training and testing on the terminal? Can you give me some information?

I want to know your strategy for retraining the network. Can you release training network code?

Hi,
Thanks for sharing the good work. Since its excellent performance, I want to use it to train and test on my own dataset, but I can not find the related doc. or codes in the project.
From the issues before, I found that you will disclose the training code after the paper publication, so how is it going? Hope to get reply soon, Thank you. :)

Your SEModule is different from what the paper claimed, why do not use fc-->fc and expand as the input x's dimension?

In the paper, the last InvertedResidual block with stride=2 is the second last; while in the code, it is the third last.

Is it a mistake in the paper or do I misunderstand something?

Hi. Thanks for your great work.
Could you please provide instructions on how to convert MoGA from Pytorch '.pth' to SNPE '.dlc'?
Is the Hswish layer supported by Qualcomm?
Did you first convert it to ONNX and then transfer it to dlc?