FGCNet: Fast Graph Convolution for Matching Features
ISMAR 2022
We provide two videos to demonstrate the effectiveness of our method for both indoor and outdoor AR/VR applications. For each video, we first reconstruct a sparse 3D points cloud using SfM with our matching method and then provide visualization of the 3D points cloud from a variety of viewpoints. Finally, given the prereconstructed 3D points cloud, we perform visual localization with our matching method. To show the stability or smoothness of using our method for visual localization, we place virtual objects into the real world and render synthetic images with estimated 6-DoF poses.
Please check our videos for more details.