1.python2,tensorboard, pytorch-0.4.0
2.put pascal-voc2012 dataset in data data ----gt (labels) ----img (images) ----train.txt (train name) ----val.txt (val name) (if you want to train your datasets, you can put the datasets as i showed above.)
3.python train.py
4.python multi_scale_test.py
I didn't test the model 'my_deeplabv3_mit.py' in voc2012, and you can modify the model as long as you like. But another model made by me can achieve 83.6 miou in pascal-voc2012,the result is http://host.robots.ox.ac.uk:8080/anonymous/TOEAAY.html. we will upload the model soon. this code is to help you to finish your project. you can use it in a semantic segmentation project.