engineerJPark / EPS_Implementation

Implementation for EPS(Explicit Pseudo-pixel Supervision)

This is the reproduction for Explicit Pseudo-pixel Supervision (EPS), which is accepted in CVPR 2021.

Paper Link : here

EPS is method for Weakly Supervised Semantic Segmentation, and they use localization map and saliency map together for improving pseudo-label quality.

Implementation detail might differ from the original implementation.

Do this for

cd eps && bash do.sh

Dependency & TroubleShooting

for installing overall depency of this repo, follow this installation :

sudo apt update && sudo apt install git && sudo apt-get install tmux
sudo apt install build-essential && sudo apt-get remove cython && pip install -U cython && pip install git+https://github.com/lucasb-eyer/pydensecrf.git
pip install matplotlib && pip install imageio

Acknowledgements

Base code is from J.Ahn's repository. Thanks for original repository writer. Applying denseCRF and AffinityNet(part of IRN) is heavily done from original repository.