Line Segment Detection and Description Evaluation Framework

X. Lin, Y. Zhou, Y. Liu, and C. Zhu, "A Comprehensive Review of Image Line Segment Detection and Description: Taxonomies, Comparisons, and Challenges", in IEEE Transactions on Pattern Analysis and Machine Intelligence (T-PAMI), 2024.

Overview

This repository provides the official implementation and evaluation framework for the review paper "A Comprehensive Review of Image Line Segment Detection and Description: Taxonomies, Comparisons, and Challenges". The framework includes tools for evaluating line segment detection and description algorithms on various datasets. Additionally, it serves as a comprehensive resource for researchers and practitioners in the field of computer vision.

Key Features

• Evaluation of Line Segment Detection Algorithms: Supports evaluation with and without ground truth line segments.
• Evaluation of Line Segment Description Algorithms: Provides a standardized framework for assessing description algorithms.
• Comprehensive Datasets: Includes multiple datasets for diverse evaluation scenarios.

Getting Started

To run the evaluation framework, follow these steps:

1. Install Dependencies: Install mexopencv to obtain the necessary functions, such as DescriptorMatcher and RotatedRect.
2. Run Evaluation Scripts:
   • eva_detection_w_demo.m: Evaluate line segment detection algorithms with ground truth line segments.
   • eva_detection_wo_demo.m: Evaluate line segment detection algorithms without ground truth line segments.
   • eva_description_demo.m: Evaluate line segment description algorithms.

Evaluation Datasets

The framework supports evaluation on the following datasets:

Dataset	# Groups/# Images	Evaluation Type	Ground Truth	Note
HPatches	116/696	Detection & Description	N/A	Natural images with variations in illumination and viewpoint.
KADID-10k	81/10,206	Detection & Description	N/A	Images with artificial distortions (blur, color, compression, noise, etc.).
RDNIM	17/1,739	Detection & Description	N/A	Natural images with variations in light and homographic warp.
DNIM	17/1,722	Detection & Description	N/A	Natural images with variations in light.
Apollo	1,000/2,087	Detection & Description	N/A	Synthetic images with variations in light.
VGGaffine	8/48	Detection & Description	N/A	Natural images with variations in blur, viewpoint, zoom/rotation, etc.
Wireframe	462/462	Detection	Wireframe	Natural images in indoor and outdoor scenarios.
YorkUrban	102/102	Detection	Line segment	Natural images in indoor and outdoor scenarios.

Additional Resources

• Project Homepage: Visit the project homepage for detailed information, additional resources, and updates.
• Comprehensive Collection of Line Segment Detection Algorithms: here.
• Comprehensive Collection of Line Segment Description Algorithms: here.

Contact

For questions, feedback, or further assistance, please contact us at: roylin_cv@163.com.

Citation

If you use this framework in your research, please cite the following paper:

@ARTICLE{10530374,
  author={Lin, Xinyu and Zhou, Yingjie and Liu, Yipeng and Zhu, Ce},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence}, 
  title={A Comprehensive Review of Image Line Segment Detection and Description: Taxonomies, Comparisons, and Challenges}, 
  year={2024},
  volume={46},
  number={12},
  pages={8074-8093},
  keywords={Image segmentation;Reviews;Task analysis;Image edge detection;Feature extraction;Taxonomy;Motion segmentation;Line segment description;line segment detection;line segment matching;low-level feature},
  doi={10.1109/TPAMI.2024.3400881}}