AI Paper Atlas

Research map / Label-efficient and robust learning / Recognition and long-tail

Google Landmarks Dataset v2 - A Large-Scale Benchmark for Instance-Level Recognition and Retrieval

Tobias Weyand, Andre Araujo, Bingyi Cao, Jack Sim

CVPR 2020 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 288 citations (OpenAlex)

Check your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
Train ResNet101+ArcFace on GLDv2-train-clean instead of GLDv1-trainResNet101+ArcFace · Revisited Oxford and Revisited ParismAP (%): 68.9 -> 76.2; 83.4 -> 87.3; 45.3 -> 55.6; 67.2 -> 74.2Improved
Train ResNet101+ArcFace on GLDv2-train-no-tail instead of GLDv2-train-clean (ablation)ResNet101+ArcFace · Revisited Oxford and Revisited ParismAP (%): 76.2 -> 76.1; 87.3 -> 86.6; 55.6 -> 55.1; 74.2 -> 72.5Hurt
Use GLDv2-train-clean rather than GLDv1-train for ResNet101+ArcFaceGLDv2 recognition · Recognition testingμAP (%): 33.25 -> 28.56Hurt
Use GLDv2-train-clean rather than GLDv1-train for ResNet101+ArcFaceGLDv2 retrieval · Retrieval testingmAP@100 (%): 20.67 -> 25.57Improved
Add local feature spatial verification to DELG global descriptorsDELG · GLDv2 recognition testingμAP (%): 32.03 -> 58.45Improved
Add local feature spatial verification to DELG global descriptorsDELG · GLDv2 retrieval testingmAP@100 (%): 21.71 -> 24.54Improved
Use ArcFace rather than Triplet or AP lossResNet101 global descriptor · GLDv2 retrieval testingmAP@100 (%): 18.71 (AP); 18.94 (Triplet); 21.35 (CosFace); 25.57 (ArcFace)Improved
Re-annotate missing recognition labelsGLDv2 recognition challenge methods · GLDv2 recognition testingμAP (%): 35.54 -> 69.39; 37.61 -> 66.53; 35.99 -> 53.08Improved
Re-annotate retrieval relevance labelsGLDv2 retrieval challenge methods · GLDv2 retrieval testingPrecision@100 (%): 4.73 -> 6.09; 4.83 -> 6.42; 3.97 -> 5.13Improved
Re-annotate retrieval relevance labelsGLDv2 retrieval challenge methods · GLDv2 retrieval testingmAP@100 (%): 37.25 -> 37.19; 34.75 -> 34.38; 32.18 -> 32.10No clear effect
Use GLDv2-train-clean rather than Landmarks-full or Landmarks-cleanResNet101+ArcFace · Revisited Oxford and Revisited ParismAP (%): Landmarks-full: 50.8, 70.4, 24.3, 47.1; Landmarks-clean: 54.2, 70.7, 28.3, 46.0; GLDv2-train-clean: 76.2, 87.3, 55.6, 74Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy