Research map / Detection and segmentation / Open-set and few-shot detection
Open World Entity Segmentation
Lu Qi, Jason Kuen, Yi Wang, Jiuxiang Gu, Hengshuang Zhao, Philip H. S. Torr, Zhe Lin, Jiaya Jia
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 global kernel bank (L_R^seg) to class-agnostic CondInst (ablation) | Entity segmentation framework · COCO val (ES) | AP_e^m: 28.3 -> 29.3 | Improved |
| add overlap suppression (L_o^seg) to class-agnostic CondInst (ablation) | Entity segmentation framework · COCO val (ES) | AP_e^m: 28.3 -> 29.1 | Improved |
| add both global kernel bank and overlap suppression (ablation) | Entity segmentation framework · COCO val (ES) | AP_e^m: 28.3 -> 29.8 | Improved |
| use softmax instead of sigmoid in overlap suppression loss (ablation) | Overlap suppression · COCO val (ES) | AP_e^m: 28.6 -> 29.1 | Improved |
| add sigmoid loss on top of softmax in overlap suppression (ablation) | Overlap suppression · COCO val (ES) | AP_e^m: 29.1 -> 29.0 | No clear effect |
| add static kernels in last one or two mask head layers (ablation) | Dynamic-kernel mask head · COCO val (ES) | AP_e^m: 28.3 -> 28.9 -> 29.1 | Improved |
| train all seven kernel-bank paths (ablation) | Global kernel bank · COCO val (ES) | AP_e^m: 28.3 -> 29.3 | Improved |
| use all paths at inference instead of last path (ablation) | Global kernel bank · COCO val (ES) | AP_e^m: 29.8 -> 30.0 | No clear effect |
| add the two modules under 36-epoch schedule (ablation) | Class-agnostic CondInst · COCO val (ES, 36 epochs) | AP_e^m: 30.4 -> 31.8 | Improved |
| train class-agnostic instead of class-specific | PanopticFCN · COCO val (ES, 12 epochs) | AP_e^m: 24.5 -> 26.5 | Improved |
| train on COCO+ADE20K merged without resolving label conflicts | Entity segmentation model · ADE20K val (12 epochs) | AP_e^m: 24.1 -> 26.3 | Improved |
| represent stuff by nine points instead of single center point (ablation) | Center-based stuff representation · COCO stuff | AP_e^m: 39.4 -> 40.2 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- KNet — K-Net: Towards Unified Image Segmentation (NeurIPS 2021) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- SOHES: Self-supervised Open-world Hierarchical Entity Segmentation (ICLR 2024 · similarity 0.90)
- EntitySAM: Segment Everything in Video (CVPR 2025 · similarity 0.88)
- Training-free Open-Vocabulary Semantic Segmentation via Diverse Prototype Construction and Sub-region Matching (AAAI 2025 · similarity 0.87)
- MOFI: Learning Image Representations from Noisy Entity Annotated Images (ICLR 2024 · similarity 0.87)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.85)
- USE: Universal Segment Embeddings for Open-Vocabulary Image Segmentation (CVPR 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.
