AI Paper Atlas

Research map / Detection and segmentation / CNN features

Divide&Classify: Fine-Grained Classification for City-Wide Visual Geo-Localization

Gabriele Trivigno, Gabriele Berton, Juan Aragon, Barbara Caputo, Carlo Masone

ICCV 2023 · Detection and segmentation › CNN features · approach: Local features and context in CNNs

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
replace cross-entropy linear classifiers with angular margin (AAMC) classifiers (ablation)D&C · SF-XLLR@1: 48.2 -> 61.4Improved
pipe D&C top-100 cells into CosPlace retrievalCosPlace · SF-XL test v1LR@1: 65.9 -> 71.4Improved
pipe D&C top-10 cells into NetVLAD retrievalNetVLAD · SF-XL test v1LR@1: 40.0 -> 62.6Improved
filter retrieval with HGE or CPlaNet classes instead of D&CCosPlace · SF-XL test v1LR@1: 65.9 -> 52.7 (HGE, Top-100)Hurt
use D&C top-1 cell only for retrieval filtering (ablation)CosPlace · SF-XL test v1LR@1: 65.9 -> 57.0Hurt
use ANN search (HNSW) instead of exhaustive kNN for CosPlaceCosPlace · SF-XL test v1LR@1: 64.7 -> 52.5Hurt
use more than 4 classifiers or a single group (N=1) in the class partition (ablation)D&C · SF-XLLR@1Hurt
increase cell size from 20 m to 50 m or 100 m (ablation)D&C · SF-XLLR@1Hurt
apply D&C classification to small datasets (Pitts30k, Tokyo 24/7)D&C · Pitts30k and Tokyo 24/7LR@1: LR@1 lower than 50% versus over 80% recall@1 for retrievalHurt

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