Research map / Label-efficient and robust learning / Self-supervised and contrastive
ConCL: Concept Contrastive Learning for Dense Prediction Pre-training in Pathology Images
Jiawei Yang, Hanbo Chen, Yuan Liang, Junzhou Huang, Lei He, Jianhua Yao
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 |
|---|---|---|---|
| use dense (grid-level) contrasting, DenseCL instead of MoCo-v2 | MoCo-v2 · GlaS detection | AP bb: 52.3 -> 53.9 | Improved |
| use precise concept correspondence from reference view (g-ConCL s=5) instead of learned matching | DenseCL · CRAG detection | AP bb: 52.3 -> 55.5 | Improved |
| use bootstrapped K-Means concepts (b-ConCL) instead of image-level contrast | MoCo-v2 · GlaS detection | AP bb: +4.5 AP bb and +3.1 AP mk | Improved |
| use BASNet saliency concepts despite poor masks | ConCL · GlaS detection | AP bb: 52.3 -> 56.1 | Improved |
| add concept loss (lambda 0.0 -> 1.0) (ablation) | b-ConCL · GlaS detection | AP bb: 52.3 -> 56.8 | Improved |
| add instance-loss warm-up (ablation) | b-ConCL · GlaS detection | AP bb: 56.1 -> 56.8 | Improved |
| increase number of concepts K from 1 to 8 (ablation) | b-ConCL · CRAG detection | AP bb: 50.0 -> 55.1 | Improved |
| increase K from 8 to 16 (ablation) | b-ConCL · GlaS detection | AP bb: 56.8 -> 55.7 | Hurt |
| cluster from stage f5 instead of f4 (ablation) | b-ConCL · CRAG detection | AP bb: 55.1 -> 56.5 | Improved |
| use ResNet-50 instead of ResNet-18 backbone (ablation) | b-ConCL · GlaS detection | AP bb: 56.8 -> 57.0 | No clear effect |
| use b-ConCL pre-training with C4 detector | Mask-RCNN-C4 · CRAG detection | AP bb: 49.4 -> 49.8 | Improved |
| use b-ConCL pre-training with RetinaNet | RetinaNet · GlaS detection | AP bb: 47.2 -> 52.6 | Improved |
Closest papers by meaning
- Unsupervised Foundation Model-Agnostic Slide-Level Representation Learning (CVPR 2025 · similarity 0.85)
- PathVQ: Reforming Computational Pathology Foundation Model for Whole Slide Image Analysis via Vector Quantization (NeurIPS 2025 · similarity 0.85)
- FOCUS: Knowledge-enhanced Adaptive Visual Compression for Few-shot Whole Slide Image Classification (CVPR 2025 · similarity 0.85)
- Prototype-Based Image Prompting for Weakly Supervised Histopathological Image Segmentation (CVPR 2025 · similarity 0.85)
- Nonparametric Distribution Matching for Self-Supervised Whole-Slide Image Condensation (NeurIPS 2026 · similarity 0.85)
- CPath-Omni: A Unified Multimodal Foundation Model for Patch and Whole Slide Image Analysis in Computational Pathology (CVPR 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.
