FPC-Net: Revisiting SuperPoint with Descriptor-Free Keypoint Detection via Feature Pyramids and Consistency-Based Implicit Matching

TL;DR

This paper presents FPC-Net, a novel interest point detection method that inherently associates points during detection, removing the need for descriptors and significantly reducing memory usage, with only a slight decrease in matching accuracy.

Contribution

It introduces a descriptor-free interest point detection approach using feature pyramids and consistency-based implicit matching, a departure from traditional descriptor-based methods.

Findings

Reduces memory usage by eliminating descriptors

Achieves comparable matching accuracy to traditional methods

Demonstrates effectiveness against classical and learned approaches

Abstract

The extraction and matching of interest points are fundamental to many geometric computer vision tasks. Traditionally, matching is performed by assigning descriptors to interest points and identifying correspondences based on descriptor similarity. This work introduces a technique where interest points are inherently associated during detection, eliminating the need for computing, storing, transmitting, or matching descriptors. Although the matching accuracy is marginally lower than that of conventional approaches, our method completely eliminates the need for descriptors, leading to a drastic reduction in memory usage for localization systems. We assess its effectiveness by comparing it against both classical handcrafted methods and modern learned approaches.