LIFT: Learned Invariant Feature Transform

TL;DR

LIFT is a deep learning architecture that jointly learns feature detection, orientation estimation, and description, outperforming previous methods on benchmarks without retraining.

Contribution

It introduces a unified, end-to-end trainable deep network for the entire feature point pipeline, combining detection, orientation, and description.

Findings

Outperforms state-of-the-art methods on benchmark datasets

Operates in an end-to-end differentiable manner

Does not require retraining for different datasets

Abstract

We introduce a novel Deep Network architecture that implements the full feature point handling pipeline, that is, detection, orientation estimation, and feature description. While previous works have successfully tackled each one of these problems individually, we show how to learn to do all three in a unified manner while preserving end-to-end differentiability. We then demonstrate that our Deep pipeline outperforms state-of-the-art methods on a number of benchmark datasets, without the need of retraining.