The paper http://www.robesafe.uah.es/personal/eduardo.romera/pdfs/Romera17tits.pdf
version: tensorflow >= 1.4.0 python2.7 or python >= 3.5
Due to confidentiality requirements on the project's code, I hid the pre-training model, some superparameters have been slightly modified. However if you use the default superparameters, that's OK ,you will also get a well result.
if you want to start train this code: The first you should process cityscapes datasets and imagenet 2012. You should first run 'sh train_pre_erf_net.sh' to pre-train erf_net model and then run 'sh train.sh' to fine tune model on cityscapes dataset.
trainImages.txt, trainLabels.txt, valImages.txt, valLabels.txt like these:
leftImg8bit/train/aachen/aachen_000000_000019_leftImg8bit.png
gtFine/train/aachen/aachen_000000_000019_gtFine_labelTrainIds.png
leftImg8bit/val/frankfurt/frankfurt_000000_000294_leftImg8bit.png
gtFine/val/frankfurt/frankfurt_000000_000294_gtFine_labelTrainIds.png
And you will run 'sh infer.sh' to inference the model, get visible results
any questions,please touch me : [email protected]