PaperSummary11 : MP-PolarMask

Poonam Saini
1 min readJan 11, 2025

--

The paper proposes MP-PolarMask, an extension of PolarMask using multiple auxiliary polar systems, to better handle complex shapes like concave obejcts. It enhances the original PolarMask by introducing multiple polar systems to improve mask representation for concave objects and by addressing the inefficiencies in a ray regression.

The main steps of the method are:

  1. MP-Head: It adds auxiliary polar centers in four quadrants around the main center. It uses additional rays from these auxiliary centers for finer mask prediction.
  2. MP-Assembly: It combines main and auxiliary center predictions to form the final object mask. It employs refined sequences of points and angles to improve mask quality.
  3. Loss functions: The method extends PolarMask’s losses to include auxiliary center related terms before training.

Overall, the paper achieves significant increases in precision metrics over PolarMask and segmentation performance especially for concave objects (e.g. food shapes). However it increases complexity in mask assembly and auxiliary center determination.

References:

--

--

Poonam Saini
Poonam Saini

Written by Poonam Saini

PhD Student, Research Associate @ Ulm University

No responses yet