The official implementation of "ISIM: Iterative Self-Improved Model for Weakly Supervised Segmentation".

• In Review for IJCV
• Please cite our paper if the code is helpful to your research. arxiv

• Python 3.7.11, PyTorch 1.7.0, and more in requirements.txt
• pydensecrf
• CUDA 11.5
• RTX 3090 GPU, RTX 2080 GPU (x4)

pip install -r requirements.txt
pip install git+https://github.com/lucasb-eyer/pydensecrf.git

Follow instructions in https://host.robots.ox.ac.uk/pascal/VOC/voc2012/#devkit

./scripts/experiments.sh

./scripts/infer_experiments.sh

./scripts/eval_experiments.sh

Qualitative segmentation results on the PASCAL VOC 2012 validation set.

• Release the final masks by our models. (SOON)

For any issues, please contact Cenk Bircanoglu, [email protected]

• In MID-2021, Project started.
• 15 Nov 2021, paper submitted to IJCV.
• On 21 January 2022, Got a refusal from IJCV.
• On 25 January 2022, Cenk Bircanoglu got the Covid-19 Vaccine and had severe health problems.
• On 20 November 2022, Health issues were relatively possible to manage, and we decided to continue this project.
• On 21 November 2022, Found out RecurSeed and EdgePredictMix. A similar idea (honestly, with a lot of improvements) has already been published.
• On 23 November 2022, Decided to put the paper to arxiv, share the code, note the project in history, and start from scratch with a new idea.