Research map / Detection and segmentation / Open-set and few-shot detection
CROSS: Cascaded Distillation and Dual-Constraint Grounding for Remote Sensing Referring Segmentation
Tingzhang Luo, Ruizhong Liu, Yichao Liu, Cheng Fan, Yu Liu, Jianyuan Guo
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 Cascaded Representation Extraction (CRE) distilling SAM geometry into multiple SigLIP layers (ablation) | SigLIP 2 + SAM 2 baseline · RefSegRS test | cIoU: 80.82 -> 82.55 | Improved |
| add Text-Guided Relational Distillation (TGD) on top of CRE (ablation) | CROSS · RefSegRS test | cIoU: 82.55 -> 83.15 | Improved |
| add PSCL on top of LGCD (ablation) | CROSS · RRSIS-D test | cIoU: 79.02 -> 79.89 | Improved |
| add PSCL on top of LGCD (ablation) | CROSS · RefSegRS test | cIoU: 83.15 -> 83.25 | Improved |
| add PSCL alone to baseline (ablation) | SigLIP 2 + SAM 2 baseline · RefSegRS test | cIoU: 80.82 -> 82.11 | Improved |
| use spatial counterfactual negatives only (no perspective negatives) (ablation) | PSCL · RRSIS-D test | cIoU: 79.02 -> 78.70 | Hurt |
| add perspective hard negatives to PSCL (ablation) | PSCL · RRSIS-D test | cIoU: 79.02 -> 79.34 | Improved |
| add spatial counterfactual negatives on RefSegRS (ablation) | PSCL · RefSegRS test | cIoU: 83.15 -> 83.20 | No clear effect |
| choose distillation layers {5,15,27} instead of {9,18,27} (ablation) | CROSS · RefSegRS test | cIoU: 83.25 -> 82.76 | Hurt |
| replace Qwen2.5-7B counterfactual generator with GPT-4o or Llama-3.1-8B (ablation) | PSCL · RRSIS-D test | cIoU: 79.89 -> 79.86 (GPT-4o), 79.84 (Llama) | No clear effect |
| CROSS vs RSRefSeg-2 on RRSIS-D | CROSS · RRSIS-D test | gIoU: 69.17 -> 68.92 | Mixed |
| CROSS vs RSRefSeg-2 on RefSegRS | CROSS · RefSegRS test | Pr@0.9: 34.40 -> 41.32 | 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.
- RS2-SAM 2 — RS2-SAM2: Customized SAM2 for Referring Remote Sensing Image Segmentation (AAAI 2026) · better on 2 of 2 shared results
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
- CrossEarth: Geospatial Vision Foundation Model for Domain Generalizable Remote Sensing Semantic Segmentation (TPAMI 2025 · similarity 0.90)
- In-Context Learning for Remote Sensing Vision: A Semantic-Aware Rotation-Robust Diffusion Framework (NeurIPS 2026 · similarity 0.88)
- CrossEarth-Gate: Fisher-Guided Adaptive Tuning Engine for Efficient Adaptation of Cross-Domain Remote Sensing Semantic Segmentation (CVPR 2026 · similarity 0.88)
- SemStereo: Semantic-Constrained Stereo Matching Network for Remote Sensing (AAAI 2025 · similarity 0.88)
- Towards Open-Vocabulary Remote Sensing Image Semantic Segmentation (AAAI 2025 · similarity 0.87)
- GeoCoT: Towards Reliable Remote Sensing Reasoning with Manifold Perspective (CVPR 2026 · 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.
