How to estimate homography for RANSAC and DLT ?

magicleap / SuperGluePretrainedNetwork

SuperGlue: Learning Feature Matching with Graph Neural Networks (CVPR 2020, Oral)

Other

3.31k stars 672 forks source link

How to estimate homography for RANSAC and DLT ? #111

Closed akif-caglar closed 2 years ago

akif-caglar commented 2 years ago

Hi, first of all thank you for your amazing work.

I wanted to know how did you estimate homography with RANSAC and DLT for the results at Table 1 of your paper ? Have you used cv::findHomography() function for both ? I could not find it in your source code.

Thank you very much.

sarlinpe commented 2 years ago

Homography estimation with RANSAC is indeed based on cv2.findHomography. For DLT, we implemented a weighted version in a few lines of numpy code. For reference, here is a weighted 8-point algorithm for essential matrix estimation: https://github.com/Skydes/SuperGluePretrainedNetwork/blob/master/models/utils.py#L297-L316