This is a project which just move the EM-Adapt to tensorflow. The EM-Adapt is referring to the approach for weakly-supervised semantic segmentation in the paper "Weakly- and semi- supervised learning of a DCNN for semantic image segmentation". And here, I just use the tensorflow to implement the approach with the help of the published code.
If you find this code useful in your research, please consider citing them:
@inproceedings{papandreou15weak,
title={Weakly- and Semi- Supervised Leaning of a DCNN for Semantic Image Segmentation},
author={George, Papandreou and Liang-Chieh Chen and Kevin Murphy and Alan L Yuille},
journal={arxiv:1502,02734},
year={2015}
}
for using this code, you have to do something else:
- for pascal data, please referring to its official website and to the augmental SBD data. Just download it and extract it in the ./, then run convert.py using 'python convert.py'.
- for the init.model, please referring to EM-ADAPT or google driver. And download it and extract in the model/ .
For more details, you can referring to the correspond code files or leave a message in the issue.
then, you just input the following sentence to train it.
python deeplab.py <gpu_id>
the final result on the validation dataset of pascal voc 2012 is 37.98% miou while it is 38.2% in the paper. Note that we use the crf while test the trained model, and you can look through my other project to see how to perform densecrf using python.
I just release a project to provide the code for evaluation.