Learning-Deep-Learning

Human-Centric Efficiency Improvements in Image Annotation for Autonomous Driving

June 2020

tl;dr: How to generate masks with few vertices that is easy to edit by human annotator.

Overall impression

Use extreme point clicking and DEXTR: Deep Extreme Cut CVPR 2018 (code) to improve efficiency.

The task improved human annotation of semantic segmentation by almost 4x without degradation of quality.

Key ideas

Technical details

Notes