LineSegmentsDetection

Including classical image processing and deep learning based methods

Overview of the homepage (current page)

1 The directories with a name prefix g_ show the global data.

• Each of the other directories represents one LSD method.

2 The related papers are in the g_papers directory.

• The file with a name prefix DL_ implies that DL-based method has been implemented.

Classical image processing

1-1 CannyLines 2015 （source code: C/C++）

• Model 1: parameter free
• Model 2: CannyLines v3

1-2 LSM 2016 （source code: C/C++ & Matlab）

2 MCMLSD 2017 [CVPR] （source code: C/C++ & Matlab）

Deep learning

• Source code: PyTorch

Comparison of the LSD methods (metric: sAP10):

1 L-CNN: End-to-End Wireframe Parsing ICCV2019

• github repo
• The previous state-of-the-art method
• Introduced metrics:
  • Structural mAP
  • Junction mAP

2 HAWP: Holistically-Attracted Wireframe Parsing CVPR 2020

• github repo

3 HT-HAWP: Deep Hough-Transform Line Priors ECCV 2020

• github repo

4 F-Clip: Fully Convolutional Line Parsing 2021.04

• github repo

Plus: LETR: Line Segment Detection Using Transformers without Edges CVPR 2021

• github repo
• Stage 1: 40M parameters
• Stage 2: by using 24G GPU