search

DingXiaoH / RepVGG

RepVGG: Making VGG-style ConvNets Great Again
MIT License
3.3k stars 433 forks source link

drop mIOU when convert to int8 TensorRT #46

Open TranThanh96 opened 3 years ago

TranThanh96 commented 3 years ago

I use RepVGG A2 as Backbone segmentation. After converting to engine TensorRT, float 16 keep the mIOU as the pytorch model, but when I convert to int8 TensorRT, mIOU drop ~5%. Did you try it yet?

sisrfeng commented 2 years ago

Hi, what is the speed of your models:

  1. before converting to VGG?
  2. after converting to VGG and?
  3. after converting to TensorRT? Thx~