Is the separable_conv is better than standard conv?

[yr-zh]

Hello! I train moblenet-deeplabv3+ and mobilenet-deeplabv3+ with --separable_conv open and find that latter is better than former by 1.8% (MIoU). So I want your result and I guess it is because the reducing of overfitting? Thanks a lot!

[VainF]

Hi @yiiiruu , I guess this depends on the initialization and hyperparameters. In my previous experiments, they provide similar mIoU on pascal voc. However separable_conv is slightly faster than conventional convolution.

Reference:

In this work, we refer the resulting convolution as atrous separable convolution, and found that atrous separable convolution significantly reduces the computation complexity of proposed model while maintaining similar (or better) performance.

[yr-zh]

Hi @yiiiruu , I guess this depends on the initialization and hyperparameters. In my previous experiments, they provide similar mIoU on pascal voc. However separable_conv is slightly faster than conventional convolution.

Reference:

In this work, we refer the resulting convolution as atrous separable convolution, and found that atrous separable convolution significantly reduces the computation complexity of proposed model while maintaining similar (or better) performance.

Thanks for your reply!I will try it again.