why the conv10 in SqueezeNet1.0 has a pad of 1?

forresti / SqueezeNet

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters

BSD 2-Clause "Simplified" License

2.17k stars 723 forks source link

Closed cc946079647 closed 8 years ago

cc946079647 commented 8 years ago

conv10 is a layer which has only 1*1 conv filters,could anyone please tell me why the pad is set to 1?thanks

forresti commented 8 years ago

We generated the SqueezeNet files programmatically (not writing prototxt by hand), so benign quirks like this sometimes creep in.

If you try disabling the conv10 padding, I would anticipate that it will work ok. Feel free to give it a try and let us know what happens!