Semi-Supervised Semantic Image Segmentation With Self-Correcting Networks. Mostafa S. Ibrahim, Arash Vahdat, Mani Ranjbar, William G. Macready. IEEE Computer Vision and Pattern Recognition 2020

Contents

0. Code
1. Model
2. Links
3. Citation

Code

• I am still trying to get permission from the Company, Dwave
• Our code is based on this deeplabv3+ repo. Our changes mainly:
• Extention to their loss function with extra Cross Entropy calls for the self-correction modules
• A small cnn network for merging logits
• A small cnn to encode the bbounding box and inject in the decoder at some scale (Deeplab uses 2 scales)
• Several scripts to prepare the data to be saved/loaded from the ancilary model

Model

Links

• 15-min video presentation in English / Arabic
• For detailed Slides / Slides pdf

Citation

@inproceedings{msibrahiCVPR20segmentcorrectionnet,
  author    = {Mostafa S. Ibrahim and Arash Vahdat and Mani Ranjbar and William G. Macready},
  title     = {Semi-Supervised Semantic Image Segmentation With Self-Correcting Networks.},
  booktitle = {2020 IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2020}
}