ZiweiWangTHU / BiDet

This is the official pytorch implementation for paper: BiDet: An Efficient Binarized Object Detector, which is accepted by CVPR2020.
MIT License
173 stars 34 forks source link

About xnor-bitcount implementation #4

Closed ztianlin closed 4 years ago

ztianlin commented 4 years ago

Hi, really appreciated for your excellent work.

Like many other open-source binary quantization repositories, I notice that you conduct BinarizeConv2d based on torch.nn.functional.conv2d. For training everything is good. But for inference, it seems that the absence of xnor-bitcount based convolution keeps this excellent work from extreme superiority.

Have you implemented this or have you intended to do so? Thanks very much.

Wuziyi616 commented 4 years ago

Hi! Thank you for your interest on our work. Actually our work doesn't focus on implementing 'real' binary neural networks on hardwares, we just propose a better way to train it. For the implementation of binary operations like xnor and bitcount, I recommend you to this repo which has inference time tested for binary neural networks on mobile devices.

ztianlin commented 4 years ago

Great! Thanks again ^_^