RISG: A rotation invariant SuperGlue algorithm.

Overview

A rotation invariant SuperGlue matching algorithm for cross modality image also see here, https://gitee.com/ssacn/RISG-image-matching

Some results：

• 多时相谷歌地球影像，Optical-optical

• 近红外与光学图像 near-infrared - optical images

• SAR和光学图像，sar-optical

• 光学图像和夜光图像，optical- night light

• 地图与光学图像，map - optical

• 光学图像与激光雷达深度图，optical -lidar depth

Getting start:

Python 3.7+ is recommended for running our code. Conda can be used to install the required packages:

Dependencies

• PyTorch-GPU 1.10.0+
• OpenCV
• SciPy
• Matplotlib
• pyymal
• pickle

Dataset

We collected a set of test data, including images from space-borne SAR and visible light sensors, drone thermal infrared sensors, and Google Earth images. You may find them in the directory "test" in this repository.

Usage

just for test

risgmatching.py contains the majority of the code. Run test_risg.py for testing:

python3 test_risg.py

Using RISG in your code

    with open('./config.yaml', 'r') as f:
        config = yaml.safe_load(f)
        risg = RISGMatcher(config)

        img_filename0 = 'test/01/pair1.jpg'
        img_filename1 = 'test/01/pair2.jpg'

        img0 = cv2.imread(img_filename0)
        img1 = cv2.imread(img_filename1)
        # rotate is number of directions
        mkpts0, mkpts1, conf, main_dir = risg.match(img0,img1,nrotate = 5)