Closed pGit1 closed 6 years ago
Hi,
I think the grouped convolution is still a bottleneck. The training took more time than the training with the MobileNet.
If I find some free minutes I will measure the inference time for the MobileNet and the ShuffleNet. I will update the Readme if I have the values.
The following URL contains the Tensorflow benchmark values for the MobileNet 0.75 compared to the ShuffleNet 1.5X.
https://gist.github.com/scheckmedia/6e93dd1eb45ef47eb870dccfe473d319
It seems that the ShuffleNet has a better inference time than the MobileNet.
Thanks @scheckmedia !
@scheckmedia ,
How fast is your grouped Convolution code? I've seen the lambda implementation elsewhere and it was pretty slow.
Do you have similar results?
Thanks for this great implementation