TP-LSD

Official implementation of paper "TP-LSD: Tri-points based line segment detector".

Introduction

This demo uses TP-LSD to detect line segments in an image. The repo contains two core files: (1) PyTorch weight files and (2) The Network used in TP-LSD.

Full paper PDF: TP-LSD: Tri-Points Based Line Segment Detector

Presentation PDF: Poster at ECCV 2020

Authors: Siyu Huang, Fangbo Qin, Pengfei Xiong, Ning Ding, Yijia He, Xiao Liu

Here are the line detection and tracking results running on the ICL-NUIM dataset.

Dependencies

1. DCNV2: See modeling/DCNv2/README.md
2. lbdmod: See lbdmod/README.md
3. pytorch ≥ 1.0
4. opencv-python

Running the Tracking Demo

This demo will run the TP-LSD network on an image sequence and extract line segments from the images.
The tracks are formed by the LineTracker class which finds sequential pair-wise nearest neighbors using two-way matching of the lines' descriptors. The demo script uses a helper class called VideoStreamer which can process inputs from three different input streams:

1. A directory of images, such as .png or .jpg
2. A video file, such as .mp4 or .avi
3. A USB Webcam

Additional useful command line parameters

• Use --method to specify the method of line segments detection (default: lsd, option: lsd, edlines, tplsd).
• Use --model to specify choose the model to use (default: tp320, option: tp320, tplite, tp512, hg).
• Use --skip to skip the first few images if the input is movie or directory (default: 1).
• Use --img_glob to change the image file extension (default: *.png).
• Press the q key to quit.

Run the demo on provided directory of images

python demo_line.py imgdir_path --method=tplsd

Run the demo on .mp4 file

python demo_line.py xx.mp4 --method=tplsd

Run the demo via webcam (id #1)

python demo_line.py camera --camid=1 --method=tplsd

Running the Inference Demo

Prepare a directory of consequent images, such as .png or .jpg

Hyper-parameter configurations

The basic configuration files for testing are saved in the config/test_config.py.

Inference with pretrained models

The pretrained models for TP-LSD-Res34(320), TP-LSD-Res34(512), TP-LSD-Lite(320), TP-LSD-HG(512) can be downloaded from this link. Please place the weights into <TP_LSD_root>/pretraineds.

• For testing, please run the following commad, you could modify the model used and the path of test images in config/test_config.yaml.

• python test.py

Metrics

Line Matching Average Precision

LAP metrics is proposed based on line matching score (LMS), which measure the matching degree of line segments from a camera model perspective. See README.md in Metrics/LAP.

Pixel based Metric

The pixel based metric is to compare the detected line segment with the ground truth in a pixel-wise manner. See README.md in Metrics/PixelBased.

Structural Average Precision

sAP metrics uses the sum of squared error (LMS) between the predicted end-points and their ground truths as evaluation metric. See README.md in Metrics/SAP.