leaderj1001 / mixed-depthwise-convolutional-kernels Goto Github PK

Hello, thanks for the repo! Are you able to train with this repo to similar results as the published paper?

Hi @leaderj1001 ,
I have an error when i tested your model with input_size is an even multiple of 32 such as (512 x 512, 448 x 448) but don't have this error with input_size is an odd multiple of 32 like 224x 224, 416 x 416.
How can I fix that ?

I tried your code but i have an error:

and when i summary your model with summaryX i get an error

How can I fix it ?

Hello,

Have you trained MixNet models for Cifar-10 from scratch (not transfer learning from ImageNet models)?

With default configuration, I obtain about 70% accuracy for MixNet small, and the accuracy decreases as the model size increases. Do I need to change some particular parameters in the configuration while training Cifar-10?

Thank you.