AI Paper Atlas

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

Stochastic Attraction-Repulsion Embedding for Large Scale Image Localization

Liu Liu, Hongdong Li, Yuchao Dai

ICCV 2019 · Label-efficient and robust learning › Recognition and long-tail · approach: Correspondence and causal representation learning · 116 citations (OpenAlex)

Read the paperCheck 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 NetVLAD with Gaussian SARE (Our-Ind.) instead of standard NetVLAD trainingNetVLAD · Sf-0Recall@1 (%): 75.58 -> 80.60Improved
train NetVLAD with Gaussian SARE (Our-Ind.) instead of CRN's bestNetVLAD · 24/7 TokyoRecall@1 (%): 75.20 -> 79.68Improved
train with Gaussian SARE (Our-Ind.) instead of standard NetVLAD trainingNetVLAD · Pitts250k-testRecall@1 (%): 85.95 -> 88.97Improved
train with Gaussian SARE (Our-Joint) instead of standard NetVLAD trainingNetVLAD · TokyoTM-valRecall@1 (%): 93.85 -> 94.71Improved
handle multiple negatives jointly instead of independently (ablation)SARE · Pitts250k-testRecall@1 (%): 88.43 vs 88.97No clear effect
handle multiple negatives jointly instead of independently (ablation)SARE · 24/7 TokyoRecall@1 (%): 79.68 vs 80.63No clear effect
handle multiple negatives jointly instead of independently (ablation)SARE · Sf-0Recall@1 (%): 80.60 -> 77.75Hurt
use Gaussian kernel instead of Cauchy or Exponential kernel in SARE (ablation)SARE · Pitts250k-test, TokyoTM-val, 24/7 Tokyo, Sf-0RecallImproved
use Gaussian SARE (Our-Ind.) instead of Contrastive losscontrastive loss · Sf-0Recall@1 (%): 74.63 -> 80.60Improved
use Gaussian SARE (Our-Ind.) instead of Geo-classification lossGeo-classification loss · 24/7 TokyoRecall@1 (%): 71.43 -> 79.68Improved
use Gaussian SARE (Our-Ind.) instead of N-pair / Angular / Lifted / Ratio lossmetric learning losses · Pitts250k-test, TokyoTM-val, 24/7 Tokyo, Sf-0Recall@1 (%): Our-Ind. 88.97 vs Angular 88.60 on Pitts250k-testNo clear effect
train with SARE (Our-Ind.) and test on image retrieval without fine-tuningNetVLAD · Oxford 5k (full)mAP: 69.09 -> 71.66Improved

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