PIXER
Learning visual information utility for feature selection and more reliable visual odometry
PIXER models the utility and uncertainty of visual information before a downstream feature extractor runs. The result is a featureness-driven pipeline that improves visual odometry accuracy while using fewer features.