Research map / Detection and segmentation / Open-set and few-shot detection
UniGeoSeg: Towards Unified Open-World Segmentation for Geospatial Scenes
Shuo Ni, Di Wang, He Chen, Haonan Guo, Ning Zhang, Jing Zhang
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 TATE and LKM to the baseline (ablation) | UniGeoSeg · Reasoning (GeoSeg-Bench, ablation subset) | gIoU: 32.88 -> 37.06 | Improved |
| add TATE (task-adaptive text enhancement) (ablation) | UniGeoSeg · SIOR interactive (ablation subset) | gIoU: 81.97 -> 84.84 | Improved |
| add LKM (latent knowledge memory) (ablation) | UniGeoSeg · Referring (GeoSeg-Bench, ablation subset) | gIoU: 64.70 -> 66.37 | Improved |
| add LKM alone without TATE (ablation) | UniGeoSeg · SIOR interactive (ablation subset) | gIoU: 82.51 -> 81.97 | Hurt |
| use task-specific TATE branches instead of one reasoning branch for all tasks (ablation) | UniGeoSeg · SIOR interactive (ablation subset) | gIoU: 83.16 -> 84.84 | Improved |
| use task-specific TATE branches instead of one referring branch for all tasks (ablation) | UniGeoSeg · SIOR interactive (ablation subset) | gIoU: 81.29 -> 84.84 | Improved |
| set LKM memory slots N=4 and lambda=0.2 (ablation) | UniGeoSeg · Reasoning (GeoSeg-Bench, ablation subset) | gIoU: 35.64 (lambda=0) -> 37.06 | Improved |
| raise LKM fusion weight lambda from 0.2 to 0.3 (N=4) (ablation) | UniGeoSeg · Reasoning (GeoSeg-Bench, ablation subset) | gIoU: 37.06 -> 35.47 | Hurt |
| use progressive task scheduling instead of average sampling (ablation) | UniGeoSeg · Reasoning (GeoSeg-Bench) | gIoU: 51.11 -> 51.42 | No clear effect |
| finetune baselines on GeoSeg-1M (ablation) | Segearth-R1 · GeoSeg-Bench Reasoning | gIoU: 11.97 -> 51.56 | Improved |
| use UniGeoSeg instead of SAM2 on box prompts | UniGeoSeg · SIOR zero-shot interactive (box) | gIoU: 92.27 (SAM2) vs 90.94 | Hurt |
| use UniGeoSeg instead of Geopixel on RRSIS-D | UniGeoSeg · RRSIS-D | cIoU: Geopixel 81.77 vs 79.78 cIoU; gIoU 67.99 vs 69.25 | Mixed |
Closest papers by meaning
- GeoViS: Geospatially Rewarded Visual Search for Remote Sensing Visual Grounding (CVPR 2026 · similarity 0.89)
- InstructSAM: A Training-free Framework for Instruction-Oriented Remote Sensing Object Recognition (NeurIPS 2025 · similarity 0.89)
- LISAt: Language-Instructed Segmentation Assistant for Satellite Imagery (NeurIPS 2025 · similarity 0.89)
- Towards Open-Vocabulary Remote Sensing Image Semantic Segmentation (AAAI 2025 · similarity 0.89)
- ROS-SAM: High-Quality Interactive Segmentation for Remote Sensing Moving Object (CVPR 2025 · similarity 0.88)
- Towards Faithful Reasoning in Remote Sensing: A Perceptually-Grounded GeoSpatial Chain-of-Thought for Vision-Language Models (ICLR 2026 · 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.
