AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Efficient Neighbourhood Consensus Networks via Submanifold Sparse Convolutions

Ignacio Rocco, Relja Arandjelović, Josef Sivic

ECCV 2020 · Architectures and efficiency › CNN design · approach: Dense matching and estimation CNNs · 189 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
use Sparse-NCNet (sparse 4D correlation with submanifold sparse convolutions) instead of dense NCNetNCNet · HPatches Sequencestime, GPU memory, MMA: up to 10x faster and up to 20x less memoryImproved
use Sparse-NCNet instead of NCNet, both with hard relocalisationSparse-NCNet · InLocspeed and memory: almost 7x faster and 6.5x less memoryImproved
add soft relocalisation after hard relocalisation (H+S vs H) (ablation)Sparse-NCNet · HPatches SequencesMMAImproved
increase feature resolution to 200x150 (enabled by sparsity) (ablation)Sparse-NCNet · HPatches SequencesMMA, speed, memory: 30% faster and 3x more memory efficient than best NCNet variantImproved
use two-stage relocalisation (H+S) at 200x150 resolution (ablation)Sparse-NCNet · InLoclocalisation accuracyImproved
use Sparse-NCNet instead of D2-Net / R2D2Sparse-NCNet · Aachen Day-Nightcorrectly localised queries: 68.4 vs 66.3 (1.0m, 5deg); 86.7 vs 88.8 (5.0m, 10deg)Mixed

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