train

chmod 755 train.sh
./train.sh

test

chmod 755 test.sh
./test.sh

result

run

python ./fcn/train_net.py --cfg ./experiments/cfgs/fcn_nlpr.yml --network FCN_32s  --iters 300000 --restore 0 --pre_train=./data/VGGnet_imagenet.npy
python ./lib/utils/converckpt2npy.py   to get `32s.npy`
python ./fcn/train_net.py --cfg ./experiments/cfgs/fcn_nlpr.yml --network FCN_16s  --iters 300000 --restore 0 --pre_train=./data/32s.npy

then train FCN_8s

setup

Requirements: software

1. tensorflow ,Python packages: cython, python-opencv, easydict (recommend to install: Anaconda)

2. link the preprocessing modeule ln -sf tensorflow/models/slim/preprocessing ./lib/fcn/utils/preprocessing