Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Mirror-Yolo: A Novel Attention Focus, Instance Segmentation and Mirror Detection Model

Fengze Li, Jieming Ma, Zhongbei Tian, Ji Ge, Hai-Ning Liang, Yungang Zhang, Tianxi Wen

TL;DR

This work proposes Mirror-YOLO, which targets mirror detection, containing a novel attention focus mechanism for features acquisition, a hypercolumn-stairstep approach to better fusion the feature maps, and the mirror bounding polygons for instance segmentation.

Abstract

Mirrors can degrade the performance of computer vision models, but research into detecting them is in the preliminary phase. YOLOv4 achieves phenomenal results in terms of object detection accuracy and speed, but it still fails in detecting mirrors. Thus, we propose Mirror-YOLO, which targets mirror detection, containing a novel attention focus mechanism for features acquisition, a hypercolumn-stairstep approach to better fusion the feature maps, and the mirror bounding polygons for instance segmentation. Compared to the existing mirror detection networks and YOLO series, our proposed network achieves superior performance in average accuracy on our proposed mirror dataset and another state-of-art mirror dataset, which demonstrates the validity and effectiveness of Mirror-YOLO.

Mirror-Yolo: A Novel Attention Focus, Instance Segmentation and Mirror Detection Model

TL;DR

This work proposes Mirror-YOLO, which targets mirror detection, containing a novel attention focus mechanism for features acquisition, a hypercolumn-stairstep approach to better fusion the feature maps, and the mirror bounding polygons for instance segmentation.

Abstract

Mirrors can degrade the performance of computer vision models, but research into detecting them is in the preliminary phase. YOLOv4 achieves phenomenal results in terms of object detection accuracy and speed, but it still fails in detecting mirrors. Thus, we propose Mirror-YOLO, which targets mirror detection, containing a novel attention focus mechanism for features acquisition, a hypercolumn-stairstep approach to better fusion the feature maps, and the mirror bounding polygons for instance segmentation. Compared to the existing mirror detection networks and YOLO series, our proposed network achieves superior performance in average accuracy on our proposed mirror dataset and another state-of-art mirror dataset, which demonstrates the validity and effectiveness of Mirror-YOLO.
Paper Structure (9 sections, 5 equations, 4 figures, 3 tables)

This paper contains 9 sections, 5 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. The Proposed Methods
  3. Attention Focus Mechanism Module
  4. Hypercolumn-stairstep
  5. Mirror Bounding Polygons
  6. Experiments and Results
  7. Ablation Experiment
  8. Experimental Results of Different Detection Models
  9. Conclusion

Figures (4)

  • Figure 1: Problems of detecting mirrors by YOLOv4.
  • Figure 2: The network comparison of the origin YOLOv4 (top) and Mirror-YOLO (bottom).
  • Figure 3: Our implementation (a) of CBAM module compared with its counterparts. The network structure of CBAM (e) and SE (f).
  • Figure 4: Comparison between the results obtained by bounding boxes (top) and our mirror bounding polygons (bottom).