Research map / Detection and segmentation / Open-set and few-shot detection
Self-Supervised Visual Representation Learning with Semantic 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 |
|---|---|---|---|
| add group-level contrastive loss to semantic grouping loss (lambda_g 1.0 -> 0.5) (ablation) | SlotCon · PASCAL VOC | mIoU: 70.1 -> 71.6 | Helped |
| use contrastive group loss instead of BYOL-style regression (ablation) | SlotCon · COCO detection | AP: 40.7 -> 41.0 | Helped |
| apply inverse augmentation on assignments rather than projections (ablation) | SlotCon · ADE20K | mIoU: 38.0 -> 39.0 | Helped |
| add geometric augmentations (two different crops) (ablation) | SlotCon · PASCAL VOC | mIoU: 62.6 -> 71.6 | Helped |
| number of prototypes 128 vs 256 vs 1024 (ablation) | SlotCon · COCO detection | AP: 40.7 / 41.0 / 40.7 | Mixed |
| lower teacher temperature 0.07 -> 0.04 (ablation) | SlotCon · COCO detection | AP: 41.0 -> 40.4 | Hurt |
| add image-level contrastive loss (ablation) | SlotCon · COCO detection | AP: B=512: 41.0 -> 40.8; B=1024: 40.7 -> 41.1 | Mixed |
| batch size 512 vs 1024 (ablation) | SlotCon · COCO detection | AP: 41.0 -> 40.7 | Hurt |
| use ImageNet-1K 200ep pre-training instead of COCO 800ep | SlotCon · PASCAL VOC | mIoU: 71.6 -> 75.0 | Helped |
| add unlabeled2017 images to COCO pre-training (COCO+) | SlotCon · COCO detection | AP: 41.0 -> 41.8 | Helped |
Closest papers by meaning
- GroupContrast: Semantic-aware Self-supervised Representation Learning for 3D Understanding (CVPR 2024 · similarity 0.89)
- Self-Supervised Visual Representation Learning from Hierarchical Grouping (NeurIPS 2020 · similarity 0.89)
- GroupViT: Semantic Segmentation Emerges From Text Supervision (CVPR 2022 · similarity 0.88)
- Perceptual Group Tokenizer: Building Perception with Iterative Grouping (ICLR 2024 · similarity 0.85)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.85)
- Learning to Compose: Improving Object Centric Learning by Injecting Compositionality (ICLR 2024 · 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.