simple-IAM
A simple implementation of Learning Instance Activation Maps for Weakly Supervised Instance Segmentation, in CVPR 2019 (Spotlight)
A simple implementation as my homework, modified based on ultra-thin-PRM.
Implementation details With my own understanding of the paper, it may be different from the author.
This implementation performs better when training demo and other datasets with fewer samples, but when I use VOC2012 training, the loss cannot be reduced. :(
If you have any good suggestions, please let me know. Thank you !
Reference
@article{Zhu2019IAM,
title={{Learning Instance Activation Maps for Weakly Supervised Instance Segmentation}},
author = {Zhu, Y. and Zhou, Y. and Xu, H. and Ye, Q. and Doermann, D. and Jiao, J.},
booktitle = {CVPR},
year = {2019}
}