search

anilsathyan7 / Portrait-Segmentation

Real-time portrait segmentation for mobile devices
MIT License
638 stars 133 forks source link

Transpose_seg (deconv) training #16

Closed emepetres closed 4 years ago

emepetres commented 4 years ago

Hi, I'm interested in training the model models/transpose_seg/deconv_bnoptimized_munet.h5.

I've seen the model has a good ratio resources consumption vs output, but still I'd like to improve the results a little bit if possible.

However I don't see in the source code should I do it. I understand that train.py and portrait_segmentation.ipynb are for the bilinear model, and portrait_segmentation_v3.ipynb for the _mnv3seg.

How could I generate/train the deconv one?

Thanks

anilsathyan7 commented 4 years ago

If you check the portrait_segmentation.ipynb, there are two functions for convolutional blocks, i.e deconv_block_rez and deconv_block. You can simply replace one with another in the network definition function and try to train the same ...

emepetres commented 4 years ago

I see, thanks a lot!