When doing patch-based pixel inference with caffe inference time is very slow. Useful trick is to feed the whole image by transforming last fc layers to convolution layers and then interpolating between neighboring patches.
the code is here
https://github.com/choosehappy/public/tree/master/EfficientPixelWiseDL
the code is here
https://github.com/choosehappy/public/tree/master/EfficientPixelWiseDL
No comments:
Post a Comment