Research map / Detection and segmentation / Open-set and few-shot detection
Self-Supervised Visual Representation Learning from Hierarchical Grouping
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 hierarchical region-guided pixel contrastive learning instead of instance-level InstFeat | ResNet-50 pre-training on PASCAL + COCO · PASCAL val semantic segmentation | mIoU: 38.11 -> 46.51 | Helped |
| Use hierarchical region guidance and cross-image negatives instead of SegSort | ResNet-50 pre-training on PASCAL + COCO · PASCAL val semantic segmentation | mIoU: 36.15 -> 46.51 | Helped |
| Use region-guided pixel contrastive pre-training rather than MoCo for 80 epochs | ResNet-50 pre-training on PASCAL + COCO · PASCAL val semantic segmentation | mIoU: 42.07 -> 46.51 | Helped |
| Use region-guided pixel contrastive pre-training rather than MoCo on PASCAL only | ResNet-50 pre-training · PASCAL val semantic segmentation | mIoU: 32.04 -> 43.51 | Helped |
| Replace the structured-edge detector with HED (ablation) | Ours · PASCAL val semantic segmentation | mIoU: 46.51 -> 48.82 | Helped |
| Add cross-image negative sampling (ablation) | Ours · PASCAL val semantic segmentation | mIoU: 39.63 -> 45.57 | Helped |
| Add cross-image negative sampling (cns=5) (ablation) | Ours · DAVIS-2017 val instance mask tracking | J (Mean): 43.1 -> 47.1 | Helped |
| Add cross-image negative sampling (cns=5) (ablation) | Ours · DAVIS-2017 val instance mask tracking | F (Mean): 46.7 -> 48.9 | Helped |
| Add pixel-wise contrastive targets to MoCo (ablation) | MoCo on ImageNet · PASCAL val semantic segmentation at 30 epochs | mIoU: 53.84 -> 55.13 | Helped |
| Add pixel-wise contrastive targets to MoCo (ablation) | MoCo on ImageNet · PASCAL val semantic segmentation at 200 epochs | mIoU: 65.53 -> 64.70 | Hurt |
| Increase MoCo pre-training from 80 to 200 epochs | MoCo on PASCAL + COCO · PASCAL val semantic segmentation | mIoU: 42.07 -> 47.01 | Helped |
| Use Ours instead of MoCo for end-to-end fine-tuning after 80-epoch pre-training | ResNet-50 pre-training on PASCAL + COCO · PASCAL val semantic segmentation | mIoU: 46.9 vs. 47.2 | No clear effect |
Closest papers by meaning
- Self-Supervised Visual Representation Learning with Semantic Grouping (NeurIPS 2022 · similarity 0.89)
- Revisiting Self-Supervised Visual Representation Learning (CVPR 2019 · similarity 0.89)
- Improving Visual Recognition with Hyperbolical Visual Hierarchy Mapping (CVPR 2024 · similarity 0.87)
- Self-Organizing Visual Prototypes for Non-Parametric Representation Learning (ICML 2025 · similarity 0.87)
- Harnessing small projectors and multiple views for efficient vision pretraining (NeurIPS 2024 · similarity 0.86)
- GroupContrast: Semantic-aware Self-supervised Representation Learning for 3D Understanding (CVPR 2024 · similarity 0.86)
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.