Research map / Detection and segmentation / Open-set and few-shot detection
SegPrompt: Boosting Open-World Segmentation via Category-Level Prompt Learning
Muzhi Zhu, Hengtao Li, Hao Chen, Chengxiang Fan, Weian Mao, Chenchen Jing, Yifan Liu, Chunhua Shen
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 |
|---|---|---|---|
| class-agnostic training instead of class-aware (ablation) | Mask2Former · LVIS-OW | AR_all: 40.8 -> 42.5 | Improved |
| class-agnostic training on known categories (ablation) | Mask2Former · LVIS-OW | AR_known: 50.2 -> 49.9 | No clear effect |
| add SegPrompt category-level prompt auxiliary supervision | class-agnostic Mask2Former · LVIS-OW | AR_all: 42.5 -> 44.9 | Improved |
| add SegPrompt auxiliary supervision | class-agnostic Mask2Former · LVIS-OW | AR_unseen: 47.4 -> 50.3 | Improved |
| use prompt-based per-category supervision instead of vanilla Mask2Former auxiliary classification (ablation) | SegPrompt · LVIS-OW | AR_all: 43.0 -> 44.1 | Improved |
| add example supervision (ablation) | SegPrompt · LVIS-OW | AR_all: 44.1 -> 44.9 | Improved |
| add example supervision (ablation) | SegPrompt · COCO to LVIS | AP_l: 28.0 -> 24.5 | Mixed |
| add class-specific embedding s to candidate queries (ablation) | SegPrompt · LVIS-OW | AR_all: 43.3 -> 44.1 | Improved |
| add prompt p to candidate queries (ablation) | SegPrompt · LVIS-OW | AR_all: 42.8 -> 44.1 | Improved |
| use reference attention instead of shared global self-attention (ablation) | SegPrompt · LVIS-OW | AR_all: 35.7 -> 44.1 | Improved |
| make candidate queries of different classes invisible to each other (ablation) | SegPrompt RefAttn · LVIS-OW | AR_all: 44.0 -> 44.1 | No clear effect |
| apply SegPrompt in full LVIS supervised training | Mask2Former · LVIS val | AP@100: 20.1 -> 22.8 | Improved |
Closest papers by meaning
- Exploring Transformers for Open-world Instance Segmentation (ICCV 2023 · similarity 0.90)
- WOW-Seg: A Word-free Open World Segmentation Model (ICLR 2026 · similarity 0.90)
- SegGPT: Towards Segmenting Everything in Context (ICCV 2023 · similarity 0.90)
- OpenWorldSAM: Extending SAM2 for Universal Image Segmentation with Language Prompts (NeurIPS 2025 · similarity 0.89)
- The Devil is in the Object Boundary: Towards Annotation-free Instance Segmentation using Foundation Models (ICLR 2024 · similarity 0.88)
- Open-World Semantic Segmentation Including Class Similarity (CVPR 2024 · similarity 0.88)
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.
