Research map / Detection and segmentation / Weakly supervised detection and segmentation
LLM-Assisted Semantic Guidance for Sparsely Annotated Remote Sensing Object Detection
Wei Liao, Chunyan Xu, Chenxu Wang, Zhen Cui
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 Adaptive Hard-Negative Reweighting to Dense Teacher baseline (ablation) | Dense Teacher · DOTA 5% labels | mAP: 50.82 -> 61.60 | Improved |
| add LSP and CLA on top of AHR (ablation) | AHR-only · DOTA 5% labels | mAP: 61.60 -> 63.65 | Improved |
| add Multi-Branch Input architecture (ablation) | AHR+LSP+CLA · DOTA 5% labels | mAP: 63.65 -> 63.95 | Improved |
| add Multi-Branch Input architecture (ablation) | AHR+LSP+CLA · DOTA 2% labels | mAP: 58.73 -> 59.37 | Improved |
| add Multi-Branch Input architecture (ablation) | AHR+LSP+CLA · DOTA 10% labels | mAP: 64.77 -> 65.11 | Improved |
| add LLM-guided class prompt to class-aware assignment (ablation) | Ours (no-prompt) · DOTA 2% labels | mAP: 56.17 -> 58.73 | Improved |
| add LLM-guided class prompt to class-aware assignment (ablation) | Ours (no-prompt) · DOTA 5% labels | mAP: 61.97 -> 63.65 | Improved |
| replace LLM-predicted prompt with ground-truth prompt (ablation) | Ours (LLM-guided prompt) · DOTA 5% labels | mAP: 63.65 -> 65.04 | Improved |
| replace LLM-predicted prompt with ground-truth prompt (ablation) | Ours (LLM-guided prompt) · DOTA 10% labels | mAP: 64.77 -> 66.41 | Improved |
| refine LLM class prompts with 10% annotations (union with labels) (ablation) | LSP module · DOTA prompt accuracy | exact-match images: 4861 -> 8447 | Improved |
| use remote-sensing-adapted LLM instead of generic LLaVA-v1.5-7B prompts (ablation) | LSP module · DOTA prompt accuracy | error images: 8933 -> 5471 | Improved |
| vary AHR threshold and weight (ablation) | AHR · DOTA training set | mAP: 67.53 - 68.20 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- SOOD — SOOD: Towards Semi-Supervised Oriented Object Detection (CVPR 2023) · better on 4 of 4 shared results
- Dense Teacher — Dense Teacher: Dense Pseudo-Labels for Semi-Supervised Object Detection (ECCV 2022) · better on 4 of 4 shared results
- ReDet — ReDet: A Rotation-Equivariant Detector for Aerial Object Detection (CVPR 2021) · better on 1 of 1 shared result
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
- S²Teacher: Step-by-step Teacher for Sparsely Annotated Oriented Object Detection (AAAI 2026 · similarity 0.90)
- InstructSAM: A Training-free Framework for Instruction-Oriented Remote Sensing Object Recognition (NeurIPS 2025 · similarity 0.88)
- Locate Anything on Earth: Advancing Open-Vocabulary Object Detection for Remote Sensing Community (AAAI 2025 · similarity 0.86)
- GeoCoT: Towards Reliable Remote Sensing Reasoning with Manifold Perspective (CVPR 2026 · similarity 0.86)
- VLRS-Bench: A Vision-Language Reasoning Benchmark for Remote Sensing (NeurIPS 2026 · similarity 0.85)
- Remote Sensing Vision-Language Foundation Models without Annotations via Ground Remote Alignment (ICLR 2024 · similarity 0.85)
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.
