Polygon-RNN++
Polygon-RNN++ is an object instance segmentation tool that can be used to interactively annotate segmentation datasets. The model builds on top of Polygon-RNN, but introduces several important improvements in both, automatic and interactive modes.
In the fully automatic mode (no annotator in the loop), Polygon-RNN++ outperforms Polygon-RNN by 10% mean IoU on the Cityscapes dataset. In interactive mode, it requires 50% fewer clicks as compared to Polygon-RNN.
Related publications:
David Acuna, Huan Ling, Amlan Kar, Sanja Fidler, "Efficient Interactive Annotation of Segmentation Datasets with Polygon-RNN++", IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2018.
Lluis Castrejon, Kaustav Kundu, Raquel Urtasun, Sanja Fidler, "Annotating Object Instances with a Polygon-RNN", IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2017.