AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Self-Supervised Image Representation Learning With Geometric Set Consistency

Nenglun Chen, Lei Chu, Hao Pan, Yan Lu, Wenping Wang

CVPR 2022 · Detection and segmentation › Open-set and few-shot detection · approach: Unsupervised object and part discovery · 8 citations (OpenAlex)

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 geometric consistency set-InfoNCE after Pri3D View and Geo pre-trainingOurs · ScanNet semantic segmentation, ResNet50mIoU: 61.7 -> 63.1Improved
Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-trainingOurs · NYUv2 semantic segmentationmIoU: 54.7 -> 55.4Improved
Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-trainingOurs · ScanNet object detectionAP: 26.6 -> 26.9Improved
Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-trainingOurs · ScanNet instance segmentationAP: 18.7 -> 19.5Improved
Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-trainingOurs · NYUv2 object detectionAP: 19.4 -> 19.7Improved
Add geometric consistency set-InfoNCE after Pri3D View and Geo pre-trainingOurs · NYUv2 instance segmentationAP: 15.8 -> 16.5Improved
Use asymmetric set feature aggregation rather than averaging all set features (ablation)Geometric set-InfoNCE · ScanNet semantic segmentationmIoU: 63.1 vs 63.1No clear effect
Vary geometric set clustering edge threshold from 0.01 to 0.05 (ablation)Geometric consistency set pre-training · ScanNet semantic segmentationmIoU: 63.2; 62.9; 62.9; 63.2; 63.1No clear effect
Replace geometry-derived consistency sets with ground-truth semantic sets (ablation)Set-InfoNCE pre-training · ScanNet semantic segmentationmIoU: 63.1 -> 66.4Improved
Continue Pri3D training with View loss (ablation)Pri3D · ScanNet semantic segmentationmIoU: 61.7 -> 61.4Hurt
Continue Pri3D training with Geo loss (ablation)Pri3D · ScanNet semantic segmentationmIoU: 61.7 -> 61.6Hurt
Continue Pri3D training with View and Geo losses (ablation)Pri3D · ScanNet semantic segmentationmIoU: 61.7 -> 61.5Hurt

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