search

Joker316701882 / Additive-Margin-Softmax

This is the implementation of paper <Additive Margin Softmax for Face Verification>
492 stars 149 forks source link

Hi, another question #4

Closed alexliyang closed 6 years ago

alexliyang commented 6 years ago

Hi, I use MS1M as train dataset , but I can only get acc = 0.73, align is done on the dataset, imagesize = 224.

Is here any tricks?

Joker316701882 commented 6 years ago

@alexliyang Not sure what happened there. Original image size is (112,96). It works fine with resface20, but with 224 input image size, receptive field may not be large enough to see the whole face. Two solutions:

  1. Align your data with code provided in this repo, make sure size=(112,96).
  2. Or try deeper network structure which may have enough receptive field size for large input image.

By the way, make sure you clean MS1M in advance, it's too noisy. Hope it helps : )