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
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add VEU (vector embedding unit, SeVR) to consistency baseline (ablation) | COVER · SCR 25% segmentation (2D) | DSC %: 91.8 -> 92.9 | Improved |
| add VPA (vector pyramid aggregation) (ablation) | COVER · SCR 25% segmentation (2D) | DSC %: 92.9 -> 93.4 | Improved |
| add MVI (multi-vector integration) (ablation) | COVER · SCR 25% segmentation (2D) | DSC %: 93.4 -> 94.0 | Improved |
| enlarge VEU receptive field from 3x3 to 7x7 (ablation) | COVER · SCR 25% linear evaluation | DSC %: 44.9 -> 54.8 | Improved |
| enlarge VEU receptive field from 7x7 to 9x9 (ablation) | COVER · SCR 25% linear evaluation | DSC %: 54.8 -> 35.1 | Hurt |
| increase vector groups J from [4,4,4,1,1] to [12,12,12,3,3] (ablation) | COVER · SCR 25% linear evaluation | DSC %: 56.3 -> 48.6 | Hurt |
| use vector CL (COVER) instead of binary dense CL (DenseCL) | COVER · 8 downstream tasks average | average score %: 74.0 -> 84.5 | Improved |
| use COVER instead of GEMINI vector regression | COVER · 8 downstream tasks average | average score %: 83.2 -> 84.5 | Improved |
| pretrain with DenseCL instead of training from scratch | DenseCL · STOIC classification (3D) | AUC %: 72.0 -> 58.8 | Hurt |
| pretrain with VADeR instead of training from scratch | VADeR · FeTA21 segmentation (3D) | DSC %: 56.9 -> 43.1 | Hurt |
| pretrain with COVER instead of training from scratch | COVER · 8 downstream tasks average | average score %: 76.4 -> 84.5 | Improved |
| use VPA instead of direct large vector template for receptive field expansion (ablation) | COVER · computation (FLOPs) head | relative FLOPs: 0.2 | Improved |
Closest papers by meaning
- When Does Contrastive Visual Representation Learning Work? (CVPR 2022 · similarity 0.87)
- A Simple Framework for Contrastive Learning of Visual Representations (ICML 2020 · similarity 0.87)
- Contrastive learning of global and local features for medical image segmentation with limited annotations (NeurIPS 2020 · similarity 0.87)
- Efficient Visual Pretraining With Contrastive Detection (ICCV 2021 · similarity 0.87)
- Accelerating Augmentation Invariance Pretraining (NeurIPS 2024 · similarity 0.85)
- Randomized-MLP Regularization Improves Domain Adaptation and Interpretability in DINOv2 (NeurIPS 2025 · similarity 0.85)
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.
