AI Paper Atlas

Research map / Detection and segmentation / CNN features

MTLDesc: Looking Wider to Describe Better

Changwei Wang, Rongtao Xu, Yuyang Zhang, Shibiao Xu, Weiliang Meng, Bin Fan, Xiaopeng Zhang

AAAI 2022 · Detection and segmentation › CNN features · approach: Local features and context in CNNs · 31 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
add feature pyramid keypoint detection (ablation)SuperPoint our impl. · HPatches @3MMA%: 67.51 -> 69.88Improved
add AGCA with gated map rather than AGCA without gated map (ablation)Baseline + Context Augmentation · HPatches @3MMA%: 71.33 -> 72.28Improved
add AGCA (ablation)Baseline · HPatches @3MMA%: 67.51 -> 72.28Improved
add DLCA (ablation)Baseline · HPatches @3MMA%: 67.51 -> 71.25Improved
combine AGCA and DLCA (ablation)Baseline · HPatches @3MMA%: 67.51 -> 73.14Improved
apply consistent attention in descriptor optimization (ablation)Baseline + Context Augmentation · HPatches @3MMA%: 73.14 -> 74.94Improved
use attention-weighted descriptors during matching as well as optimization (ablation)Baseline + Context Augmentation + Consistent Attention Weighting · HPatches @3MMA%: 74.94 -> 78.66Improved
use attention-weighted descriptors during matching (ablation)Baseline + Context Augmentation + Consistent Attention Weighting · HPatches @3M.S.%: 50.67 -> 47.16Hurt
add consistent attention weighting (ablation)MTLDesc · HPatches @3MMA%: 73.14 -> 78.66Improved
use MTLDesc rather than DISK (8K)local feature matching · HPatches @3MMA%: 77.58 -> 78.66Improved
use MTLDesc rather than DISKlocal feature description · Aachen Day-Night v1.1localized queries (0.25m,2°): 72.8 -> 74.3Improved
use MTLDesc rather than R2D2local feature description · InLoc DUC1localized queries (0.5m): 57.1 -> 61.6Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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