Research map / Detection and segmentation / Open-set and few-shot detection
SOS: Segment Object System for Open-World Instance Segmentation With Object Priors
Christian Wilms, Tim Rolff, Maris N Hillemann, Robert Johanson, Simone Frintrop
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 |
|---|---|---|---|
| replace Grid prompts with DINO self-attention object prior for SAM (ablation) | SOS · COCO (VOC) -> COCO (non-VOC) | F1: 6.2 -> 14.3 | Improved |
| add SAM-generated pseudo annotations with Grid prior (ablation) | class-agnostic Mask R-CNN · COCO (VOC) -> COCO (non-VOC) | F1: 2.2 -> 6.2 | Improved |
| add confidence thresholding and NMS post-processing of pseudo annotations (ablation) | SOS with DINO prior · COCO (VOC) -> COCO (non-VOC) | F1: 14.3 -> 14.4 | Improved |
| use up to 10 instead of 3 pseudo annotations per image (ablation) | SOS · COCO (VOC) -> COCO (non-VOC) | F1: 13.4 -> 14.4 | Improved |
| increase pseudo annotations from 10 to 20 (ablation) | SOS · COCO (VOC) -> COCO (non-VOC) | F1: 14.4 -> 14.3 | No clear effect |
| use learned (CAM, DeepGaze) priors instead of simple ones (Spx, Contour, VOCUS2) (ablation) | SOS object prior · COCO (VOC) -> COCO (non-VOC) | F1: 9.4 vs 9.6-10.5 | No clear effect |
| use SOS instead of GGN pseudo-annotation OWIS | Mask R-CNN training · COCO (VOC) -> COCO (non-VOC) | F1: 8.4 -> 14.5 | Improved |
| use SOS instead of vanilla SAM | open-world instance segmentation · COCO (VOC) -> COCO (non-VOC) | F1: 6.7 -> 14.5 | Improved |
| use SOS instead of LDET | open-world instance segmentation · COCO -> UVO | F1: 28.5 -> 28.0 | Hurt |
| use SOS instead of GGN | open-world instance segmentation · COCO -> ADE20k | F1: 13.3 -> 17.0 | Improved |
| use SOS pseudo annotations instead of GGN | pseudo annotation quality · COCO train non-VOC | precision: 7.3 -> 19.0 | Improved |
Closest papers by meaning
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.88)
- SOHES: Self-supervised Open-world Hierarchical Entity Segmentation (ICLR 2024 · similarity 0.87)
- v-CLR: View-Consistent Learning for Open-World Instance Segmentation (CVPR 2025 · similarity 0.87)
- OpenWorldSAM: Extending SAM2 for Universal Image Segmentation with Language Prompts (NeurIPS 2025 · similarity 0.87)
- Semantic-aware SAM for Point-Prompted Instance Segmentation (CVPR 2024 · similarity 0.87)
- Open3DIS: Open-Vocabulary 3D Instance Segmentation with 2D Mask Guidance (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.
