search

Randl / MobileNetV2-pytorch

Impementation of MobileNetV2 in pytorch
https://arxiv.org/abs/1801.04381
MIT License
271 stars 83 forks source link

use the int8 #3

Closed YanShuo1992 closed 6 years ago

YanShuo1992 commented 6 years ago

I am testing the different data type. I am curious about why you put the int8 in the TODO. Is there any different between using float and int? I am new in this field. Can you give me a hint?

Randl commented 6 years ago

lower precision is potentially faster than floating point one, however pytorch doesn't support it yet.