4 months ago

Revisiting Oxford and Paris: Large-Scale Image Retrieval Benchmarking

Filip Radenović; Ahmet Iscen; Giorgos Tolias; Yannis Avrithis; Ondřej Chum

Abstract

In this paper we address issues with image retrieval benchmarking on standard and popular Oxford 5k and Paris 6k datasets. In particular, annotation errors, the size of the dataset, and the level of challenge are addressed: new annotation for both datasets is created with an extra attention to the reliability of the ground truth. Three new protocols of varying difficulty are introduced. The protocols allow fair comparison between different methods, including those using a dataset pre-processing stage. For each dataset, 15 new challenging queries are introduced. Finally, a new set of 1M hard, semi-automatically cleaned distractors is selected. An extensive comparison of the state-of-the-art methods is performed on the new benchmark. Different types of methods are evaluated, ranging from local-feature-based to modern CNN based methods. The best results are achieved by taking the best of the two worlds. Most importantly, image retrieval appears far from being solved.

Code Repositories

tensorflow/models

Mentioned in GitHub

tensorflow/models/tree/master/research/delf

Mentioned in GitHub

Benchmarks

Benchmark	Methodology	Metrics
image-retrieval-on-roxford-hard	HesAff–rSIFT–VLAD	mAP: 13.2
image-retrieval-on-roxford-hard	HesAff–rSIFT–SMK*+SP	mAP: 35.8
image-retrieval-on-roxford-hard	HesAff–rSIFT–ASMK*	mAP: 36.4
image-retrieval-on-roxford-hard	HesAff–rSIFT–HQE+SP	mAP: 49.7
image-retrieval-on-roxford-hard	HesAff–rSIFT–ASMK*+SP	mAP: 36.7
image-retrieval-on-roxford-hard	HesAff–rSIFT–SMK*	mAP: 35.4
image-retrieval-on-roxford-hard	HesAff–rSIFT–HQE	mAP: 41.3
image-retrieval-on-roxford-medium	HesAff–rSIFT–SMK*+SP	mAP: 59.8
image-retrieval-on-roxford-medium	HesAff–rSIFT–HQE+SP	mAP: 71.3
image-retrieval-on-roxford-medium	HesAff–rSIFT–ASMK*	mAP: 60.4
image-retrieval-on-roxford-medium	HesAff–rSIFT–SMK*	mAP: 59.4
image-retrieval-on-roxford-medium	HesAff–rSIFT–ASMK*+SP	mAP: 60.6
image-retrieval-on-roxford-medium	HesAff–rSIFT–VLAD	mAP: 33.9
image-retrieval-on-roxford-medium	HesAff–rSIFT–HQE	mAP: 66.3
image-retrieval-on-roxford-medium-without	HesAff–rSIFT–VLAD	Average mAP: 33.9
image-retrieval-on-rparis-hard	HesAff–rSIFT–VLAD	mAP: 17.5
image-retrieval-on-rparis-hard	HesAff–rSIFT–HQE	mAP: 44.7
image-retrieval-on-rparis-hard	HesAff–rSIFT–SMK*+SP	mAP: 31.3
image-retrieval-on-rparis-hard	HesAff–rSIFT–SMK*	mAP: 31.2
image-retrieval-on-rparis-hard	HesAff–rSIFT–ASMK*	mAP: 34.5
image-retrieval-on-rparis-hard	HesAff–rSIFT–ASMK*+SP	mAP: 35.0
image-retrieval-on-rparis-hard	HesAff–rSIFT–HQE+SP	mAP: 45.1
image-retrieval-on-rparis-medium	HesAff–rSIFT–SMK*	mAP: 59.0
image-retrieval-on-rparis-medium	HesAff–rSIFT–ASMK*	mAP: 61.2
image-retrieval-on-rparis-medium	HesAff–rSIFT–HQE+SP	mAP: 70.2
image-retrieval-on-rparis-medium	HesAff–rSIFT–SMK*+SP	mAP: 59.2
image-retrieval-on-rparis-medium	HesAff–rSIFT–HQE	mAP: 68.9
image-retrieval-on-rparis-medium	HesAff–rSIFT–VLAD	mAP: 43.6
image-retrieval-on-rparis-medium	HesAff–rSIFT–ASMK*+SP	mAP: 61.4

Build AI with AI

From idea to launch — accelerate your AI development with free AI co-coding, out-of-the-box environment and best price of GPUs.

AI Co-coding

Ready-to-use GPUs

Best Pricing

Get Started

Hyper Newsletters

Subscribe to our latest updates

We will deliver the latest updates of the week to your inbox at nine o'clock every Monday morning

Command Palette