A tensorflow implementation of Learning Deconvolution Network for Semantic Segmentation.
A modified working version of Fabian Bormann implementation.
Requires:
-Tensorflow
-numpy
-OpenCV2
Requires the PASCAL VOC 2012 dataset(stage-1 and stage-2) from the paper authors' website.Uses GPU for the convolution and deconvolution operations.A deconvolution network is learned separatelty for stage-1 and stage-2 training dataset and the final model is saved offline for later segmentation.