leaderj1001 / MobileNetV3-Pytorch

Implementing Searching for MobileNetV3 paper using Pytorch
MIT License
292 stars 70 forks source link

Confusing parm #1

Closed qd213618 closed 5 years ago

qd213618 commented 5 years ago

In your large model you missed a 1414112 layer, please add it

leaderj1001 commented 5 years ago

Thank you for your opinion. 캡처 Does it mean the last layer in the above figure?

qd213618 commented 5 years ago

sorry, I saw the missing layer, you are not missing it, but I think the mid layer output in your pic should be 112 not 160 in your code, because every time the feature accumulated the stride should be 2, and the only case when s=2 is in the last row of your pic.

leaderj1001 commented 5 years ago

This is a little confusing. Because the second layer of the picture is # out = 160. So I think the third layer of the picture should be 14, 14, 160.

qd213618 commented 5 years ago

yeap, quite confusing, this is the worst paper I read write by google. Maybe you can have a test on both 14 14 160 and 14 14 112 to tell which is the best. I do not have the dataset.

leaderj1001 commented 5 years ago

Ok, Thank you for your opinion. I'll try this !! thank you.