University of Alberta2025AI & ML
Feature-Augmented Image Stitching
Panorama stitching that replaces SIFT/ORB keypoints with learned neural features for homography estimation.
Classical stitching pipelines depend on hand-designed keypoint detectors, which degrade under low texture, repeated patterns and large viewpoint change. This implementation swaps in learned feature extraction — the Feature-Augmented Registration approach — to compute the homography, then blends into a seamless panorama.
The comparison against SIFT and ORB baselines is the point: it shows where learned features win and where the classical detectors are still perfectly adequate and considerably cheaper.
Gallery
