Research map / Detection and segmentation / Open-set and few-shot detection
Segmenting Known Objects and Unseen Unknowns without Prior Knowledge
Stefano Gasperini, Alvaro Marcos-Ramiro, Michael Schmidt, Nassir Navab, Benjamin Busam, Federico Tombari
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 a dedicated prototype head (ablation) | U3HS · Lost&Found (unseen) | PQ: 3.64 -> 7.94 | Improved |
| add relaxed embedding association (ablation) | semantic uncertainty baseline · Lost&Found (unseen) | PQ: 0.49 -> 3.64 | Improved |
| add semantic uncertainty branch without other components (ablation) | OSIS-style baseline · Lost&Found (unseen) | PQ: 1.45 -> 0.49 | Hurt |
| add majority voting for instance class (ablation) | U3HS · Cityscapes (open) | PQ: 23.94 -> 39.84 | Improved |
| add reassigning of DBSCAN outliers (ablation) | U3HS · Cityscapes (open) | PQ: 39.84 -> 41.21 | Improved |
| add reassigning of DBSCAN outliers (ablation) | U3HS · Lost&Found (unseen) | PQ: 7.85 -> 7.94 | No clear effect |
| learn semantic-aware embeddings (ablation) | U3HS · Lost&Found (unseen) | PQ: 2.33 -> 7.85 | Improved |
| use improved DPN uncertainty instead of softmax (ablation) | U3HS · Lost&Found (unseen) | PQ: 0.10 -> 7.94 | Improved |
| use improved DPN instead of original DPN (ablation) | U3HS · Lost&Found (unseen) | PQ: 2.09 -> 7.94 | Improved |
| use uncertainty plus instance-aware embeddings instead of learning void | U3HS vs OSIS · Lost&Found (unseen) | PQ: 1.45 -> 7.94 | Improved |
| switch from closed to open-set inference | U3HS · Cityscapes | PQ: 46.53 -> 41.21 | Hurt |
| use U3HS instead of Panoptic-DeepLab for known classes | Cityscapes closed · Cityscapes (closed) | PQ: 45.82 -> 46.53 | 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.
- Panoptic-DeepLab — Panoptic-DeepLab: A Simple, Strong, and Fast Baseline for Bottom-Up Panoptic Segmentation (CVPR 2020) · 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
- Unifying Training and Inference for Panoptic Segmentation (CVPR 2020 · similarity 0.89)
- Scene-Centric Unsupervised Panoptic Segmentation (CVPR 2025 · similarity 0.88)
- unMORE: Unsupervised Multi-Object Segmentation via Center-Boundary Reasoning (ICML 2025 · similarity 0.88)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.88)
- Revisiting Open-Set Panoptic Segmentation (AAAI 2024 · similarity 0.88)
- Segment Anything without Supervision (NeurIPS 2024 · 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.
