Supporting large-scale image recognition with out-of-domain samples

TL;DR

This paper introduces an efficient end-to-end method for large-scale landmark image recognition that leverages out-of-domain samples for improved filtering and ranking, achieving top results in a major challenge.

Contribution

It presents a novel approach combining CNN embeddings with out-of-domain similarity filtering for landmark recognition at scale.

Findings

Achieved 1st place in Google Landmark Recognition 2020

Effective use of out-of-domain samples for noise filtering

High accuracy in large-scale image recognition

Abstract

This article presents an efficient end-to-end method to perform instance-level recognition employed to the task of labeling and ranking landmark images. In a first step, we embed images in a high dimensional feature space using convolutional neural networks trained with an additive angular margin loss and classify images using visual similarity. We then efficiently re-rank predictions and filter noise utilizing similarity to out-of-domain images. Using this approach we achieved the 1st place in the 2020 edition of the Google Landmark Recognition challenge.