Research map / Detection and segmentation / Open-set and few-shot detection
SegEarth-R2: Towards Comprehensive Language-guided Segmentation for Remote Sensing Images
Zepeng Xin, Kaiyu Li, Luodi Chen, Wanchen Li, Xiao Yuchen, Hui Qiao, Weizhan Zhang, Deyu Meng, Xiangyong Cao
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 attention supervision (lambda_S=0.01) (ablation) | SegEarth-R2 · RRSIS-D test | gIoU: 66.6 -> 67.9 | Improved |
| add spatial attention supervision (lambda_S=0.01) (ablation) | SegEarth-R2 · RefSegRS test | gIoU: 70.4 -> 71.8 | Improved |
| add spatial attention supervision (lambda_S=0.01) (ablation) | SegEarth-R2 · EarthReason test | gIoU: 72.9 -> 73.5 | Improved |
| raise lambda_S to 0.1 (ablation) | SegEarth-R2 · EarthReason test | gIoU: 73.5 -> 71.8 | Hurt |
| use a single [SEG] query instead of 100 queries with matching (ablation) | SegEarth-R2 · RRSIS-D val | gIoU: 66.4 -> 68.8 | Improved |
| use a single [SEG] query instead of 100 queries (ablation) | SegEarth-R2 · inference time | s/image: 0.82 -> 0.54 | Improved |
| use a single [SEG] query instead of 100 queries (ablation) | SegEarth-R2 · compute | TFLOPs: 8.4 -> 6.1 | Improved |
| use Mask2Former decoder with Swin-B instead of SAM decoder with ViT-H (ablation) | SegEarth-R2 · EarthReason test | gIoU: 65.4 -> 73.5 | Improved |
| use Mask2Former decoder with Swin-B instead of SAM 2 decoder with ViT-H (ablation) | SegEarth-R2 · RefSegRS test | gIoU: 64.2 -> 71.8 | Improved |
| add spatial attention supervision | SegEarth-R2 · LaSeRS part-level | gIoU | Improved |
| use 3B model against 8B GeoPixel | SegEarth-R2-3B · LaSeRS multi-target | gIoU: 49.2 -> 38.3 | Hurt |
Closest papers by meaning
- LISAt: Language-Instructed Segmentation Assistant for Satellite Imagery (NeurIPS 2025 · similarity 0.92)
- VRSBench: A Versatile Vision-Language Benchmark Dataset for Remote Sensing Image Understanding (NeurIPS 2024 · similarity 0.89)
- Towards Open-Vocabulary Remote Sensing Image Semantic Segmentation (AAAI 2025 · similarity 0.88)
- SkySense-O: Towards Open-World Remote Sensing Interpretation with Vision-Centric Visual-Language Modeling (CVPR 2025 · similarity 0.88)
- GeoCoT: Towards Reliable Remote Sensing Reasoning with Manifold Perspective (CVPR 2026 · similarity 0.88)
- VLRS-Bench: A Vision-Language Reasoning Benchmark for Remote Sensing (NeurIPS 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.
