Research map / Detection and segmentation / Weakly supervised detection and segmentation
S2-UniSeg: Fast Universal Agglomerative Pooling for Scalable Segment Anything Without Supervision
Huihui Xu, Jin Ye, Hongqiu Wang, Changkai Ji, Jiashi Lin, Ming Hu, Ziyan Huang, Ying Chen, Chenglong Ma, Tianbin Li, Lihao Liu, Junjun He and 1 more
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 spatial similarity (omega_s=0.4) to the identify step (ablation) | UniAP · UVO val | AP: 12.4 -> 16.2 | Improved |
| use much lower thresholds (0.5-0.1) (ablation) | UniAP · UVO val | AP: 16.2 -> 15.5 | Hurt |
| use finer thresholds (0.9-0.1) (ablation) | UniAP · UVO val | AP: 16.2 -> 16.7 | Mixed |
| use 4 local crops instead of 2 (ablation) | S2-UniSeg · UVO val | AP: 16.2 -> 15.9 | No clear effect |
| use 6 local crops instead of 2 (ablation) | S2-UniSeg · UVO val | AR100: 32.1 -> 32.3 | No clear effect |
| increase QuerySD projection dimension K from 512 to 1024 (ablation) | S2-UniSeg · UVO val | AP: 16.2 -> 16.9 | Improved |
| reduce K from 512 to 128 (ablation) | S2-UniSeg · UVO val | AP: 16.2 -> 13.6 | Hurt |
| scale pretraining data from 0.4M to 2.0M SA-1B images | S2-UniSeg · COCO class-agnostic instance seg | AP: 33.6 -> 36.4 | Improved |
| scale pretraining data from 0.4M to 2.0M SA-1B images | S2-UniSeg · COCO class-agnostic instance seg | AR: 43.9 -> 46.3 | Improved |
| replace multi-stage offline pseudo-masks with fast UniAP and single-stage pretraining | S2-UniSeg vs U2Seg · COCO class-agnostic instance seg | AP: 7.3 -> 14.2 | Improved |
| use UniAP instead of MaskCut for pseudo-masks | pseudo-mask generation · COCO val2017 | time per image (s): 4.72 -> 0.045 | Improved |
| use UniAP instead of MaskCut for pseudo-masks | pseudo-mask generation · COCO val2017 | AP mask: 6.7 -> 6.2 | Hurt |
Closest papers by meaning
- Unsupervised Universal Image Segmentation (CVPR 2024 · similarity 0.90)
- Segment Anything without Supervision (NeurIPS 2024 · similarity 0.90)
- Universal Segmentation at Arbitrary Granularity with Language Instruction (CVPR 2024 · similarity 0.87)
- Integrating Low-Level Visual Cues for Enhanced Unsupervised Semantic Segmentation (AAAI 2025 · similarity 0.87)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.87)
- Bridge the Points: Graph-based Few-shot Segment Anything Semantically (NeurIPS 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.
