search

apple / ml-mobileone

This repository contains the official implementation of the research paper, "An Improved One millisecond Mobile Backbone".
Other
728 stars 61 forks source link

When we try to quantify the model, things go awry.... #6

Open WenWeiZhao opened 2 years ago

rohansaw commented 3 months ago

@WenWeiZhao did you find any insights on this or even fixed it?

In the original paper the authors report the inference speed on an edge TPU (quantized tflite), so I would expect that there should not be strong degradation of the results, however there is no infromation provided on this.