AI Paper Atlas

Research map / Label-efficient and robust learning / Self-supervised and contrastive

Vector Contrastive Learning For Pixel-Wise Pretraining In Medical Vision

Yuting He, Shuo Li

ICCV 2025 · Label-efficient and robust learning › Self-supervised and contrastive · approach: Empirical contrastive visual learning · 2 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 VEU (vector embedding unit, SeVR) to consistency baseline (ablation)COVER · SCR 25% segmentation (2D)DSC %: 91.8 -> 92.9Improved
add VPA (vector pyramid aggregation) (ablation)COVER · SCR 25% segmentation (2D)DSC %: 92.9 -> 93.4Improved
add MVI (multi-vector integration) (ablation)COVER · SCR 25% segmentation (2D)DSC %: 93.4 -> 94.0Improved
enlarge VEU receptive field from 3x3 to 7x7 (ablation)COVER · SCR 25% linear evaluationDSC %: 44.9 -> 54.8Improved
enlarge VEU receptive field from 7x7 to 9x9 (ablation)COVER · SCR 25% linear evaluationDSC %: 54.8 -> 35.1Hurt
increase vector groups J from [4,4,4,1,1] to [12,12,12,3,3] (ablation)COVER · SCR 25% linear evaluationDSC %: 56.3 -> 48.6Hurt
use vector CL (COVER) instead of binary dense CL (DenseCL)COVER · 8 downstream tasks averageaverage score %: 74.0 -> 84.5Improved
use COVER instead of GEMINI vector regressionCOVER · 8 downstream tasks averageaverage score %: 83.2 -> 84.5Improved
pretrain with DenseCL instead of training from scratchDenseCL · STOIC classification (3D)AUC %: 72.0 -> 58.8Hurt
pretrain with VADeR instead of training from scratchVADeR · FeTA21 segmentation (3D)DSC %: 56.9 -> 43.1Hurt
pretrain with COVER instead of training from scratchCOVER · 8 downstream tasks averageaverage score %: 76.4 -> 84.5Improved
use VPA instead of direct large vector template for receptive field expansion (ablation)COVER · computation (FLOPs) headrelative FLOPs: 0.2Improved

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