qinziwen / graphlinematching

The code for Robust Line Segments Matching via Graph Convolution Networks

This repository contains the PyTorch implementation of the paper Robust Line Segments Matching via Graph Convolution Networks.

Requirements

install python3.5.2,pytorch 1.1+,ninja-build:

sudo apt-get install ninja-build

Install python packages:

pip install tensorboardX scipy easydict pyyaml

Dataset

To train and eval the network, you should download Scannet, and then, you should use the code to pre-process (e.g., generate the grund truth label) the dataset. if you want to augment the dataset, install:

pip install imgaug

and then, run:

python3 aug_scannet.py

Training

To train the model(s) in the paper, run this command:

python3 train_eval.py --cfg your_yaml_path

📋Example python3 train_eval.py --cfg experiments/vgg16_scannet.yaml

Evaluation

To evaluate the model on Scannet, run:

python3 eval.py --cfg your_yaml_path

📋Example python3 eval.py --cfg experiments/vgg16_scannet.yaml

Visualization

To view the matching results, run:

python3 test.py --cfg experiments/vgg16_scannet.yaml --model_path params_last.pt --left_img test_data/000800.jpg --right_img test_data/000900.jpg --left_lines test_data/000800.txt --right_lines test_data/000900.txt

📋the pre-trained model trained on scannet will be provided when the paper is accepted. A example is: