Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MARS: Mask Attention Refinement with Sequential Quadtree Nodes for Car Damage Instance Segmentation

Teerapong Panboonyuen, Naphat Nithisopa, Panin Pienroj, Laphonchai Jirachuphun, Chaiwasut Watthanasirikrit, Naruepon Pornwiriyakul

TL;DR

This paper presents MARS (Mask Attention Refinement with Sequential quadtree nodes) for car damage instance segmentation, which represents self-attention mechanisms to draw global dependencies between the sequential quadtree node layer and quadtree transformer to recalibrate channel weights and predict highly accurate instance masks.

Abstract

Evaluating car damages from misfortune is critical to the car insurance industry. However, the accuracy is still insufficient for real-world applications since the deep learning network is not designed for car damage images as inputs, and its segmented masks are still very coarse. This paper presents MARS (Mask Attention Refinement with Sequential quadtree nodes) for car damage instance segmentation. Our MARS represents self-attention mechanisms to draw global dependencies between the sequential quadtree nodes layer and quadtree transformer to recalibrate channel weights and predict highly accurate instance masks. Our extensive experiments demonstrate that MARS outperforms state-of-the-art (SOTA) instance segmentation methods on three popular benchmarks such as Mask R-CNN [9], PointRend [13], and Mask Transfiner [12], by a large margin of +1.3 maskAP-based R50-FPN backbone and +2.3 maskAP-based R101-FPN backbone on Thai car-damage dataset. Our demos are available at https://github.com/kaopanboonyuen/MARS.

MARS: Mask Attention Refinement with Sequential Quadtree Nodes for Car Damage Instance Segmentation

TL;DR

This paper presents MARS (Mask Attention Refinement with Sequential quadtree nodes) for car damage instance segmentation, which represents self-attention mechanisms to draw global dependencies between the sequential quadtree node layer and quadtree transformer to recalibrate channel weights and predict highly accurate instance masks.

Abstract

Evaluating car damages from misfortune is critical to the car insurance industry. However, the accuracy is still insufficient for real-world applications since the deep learning network is not designed for car damage images as inputs, and its segmented masks are still very coarse. This paper presents MARS (Mask Attention Refinement with Sequential quadtree nodes) for car damage instance segmentation. Our MARS represents self-attention mechanisms to draw global dependencies between the sequential quadtree nodes layer and quadtree transformer to recalibrate channel weights and predict highly accurate instance masks. Our extensive experiments demonstrate that MARS outperforms state-of-the-art (SOTA) instance segmentation methods on three popular benchmarks such as Mask R-CNN [9], PointRend [13], and Mask Transfiner [12], by a large margin of +1.3 maskAP-based R50-FPN backbone and +2.3 maskAP-based R101-FPN backbone on Thai car-damage dataset. Our demos are available at https://github.com/kaopanboonyuen/MARS.
Paper Structure (21 sections, 35 equations, 3 figures, 2 tables)

This paper contains 21 sections, 35 equations, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed Method
  4. Experiments
  5. Experimental Setup
  6. Average Precision (AP)
  7. Precision and Recall
  8. AP50 and AP75
  9. Comparison with SOTA
  10. Average Precision (AP)
  11. Frames Per Second (FPS)
  12. Implementation Details
  13. Conclusions
  14. Appendix: Mathematical Details and Proofs
  15. Mask Attention Refinement
  16. ...and 6 more sections

Figures (3)

  • Figure 1: Instance segmentation results on the Thai car-damage validation set: a) Mask R-CNN he2017mask, b) PointRend kirillov2020pointrend, c) Mask Transfiner ke2022mask, d) MARS (Ours) using R50-FPN as the backbone. MARS demonstrates superior detail in high-frequency image regions by replacing the default mask head of Mask R-CNN (Zoom in for better view).
  • Figure 2: The framework of MARS. Our end-to-end trained network consists of semantic and instance segmentation modules.
  • Figure 3: Comparison of instance segmentation results: standard mask head (columns 3 and 4) versus MARS (right image). The MARS framework outperforms the standard approach by capturing significantly finer details around object boundaries, demonstrating its enhanced ability to refine mask predictions and better delineate intricate features. This improvement highlights MARS's superior precision and effectiveness in addressing complex segmentation challenges.