Research map / Label-efficient and robust learning / Self-supervised and contrastive
Semantic Concentration for Self-Supervised Dense Representations Learning
Peisong Wen, Qianqian Xu, Siran Dai, Runmin Cong, Qingming Huang
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 random cropping (ablation) | SimCLR · CIFAR-10 | accuracy: 53.1 -> 88.5 | Improved |
| add random cropping (ablation) | SimCLR · CIFAR-100 | accuracy: 19.5 -> 55.4 | Improved |
| add random cropping (ablation) | MoCo · CIFAR-100 | accuracy: 26.4 -> 62.5 | Improved |
| add random cropping (ablation) | SimSiam · CIFAR-100 | accuracy: 8.2 -> 60.3 | Improved |
| add color distortion (ablation) | SimCLR · CIFAR-10 | accuracy: 83.5 -> 88.5 | Improved |
| add color dropping (ablation) | SimCLR · CIFAR-10 | accuracy: 63.2 -> 83.5 | Improved |
| add color dropping (ablation) | Barlow Twins · CIFAR-100 | accuracy: 34.1 -> 50.4 | Improved |
| add proposed semantic concentration (CoTAP + OAF) fine-tuning | DINO ViT-S/16 · COCOStuff-27 linear segmentation | mIoU: 38.8 -> 50.5 | Improved |
| add proposed semantic concentration fine-tuning | iBOT ViT-B/16 · COCOStuff-27 linear segmentation | mIoU: 49.6 -> 53.9 | Improved |
| add proposed semantic concentration fine-tuning | DINO ViT-S/16 · ImageNet-1k k-NN classification | Acc@1: 74.5 -> 78.0 | Improved |
| add proposed semantic concentration fine-tuning | Mugs · ADE20k linear segmentation | mIoU: 30.8 -> 32.5 | Improved |
| add proposed semantic concentration fine-tuning | DINO · ADE20k linear segmentation | mIoU: 22.3 -> 30.2 | Improved |
Closest papers by meaning
- Near, far: Patch-ordering enhances vision foundation models' scene understanding (ICLR 2025 · similarity 0.88)
- Patch-Level Representation Learning for Self-Supervised Vision Transformers (CVPR 2022 · similarity 0.88)
- SCLIP: Rethinking Self-Attention for Dense Vision-Language Inference (ECCV 2024 · similarity 0.88)
- Exploring Structural Degradation in Dense Representations for Self-supervised Learning (NeurIPS 2025 · similarity 0.88)
- Learning Where To Learn in Cross-View Self-Supervised Learning (CVPR 2022 · similarity 0.87)
- Unsupervised Object-Level Representation Learning from Scene Images (NeurIPS 2021 · similarity 0.87)
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.
