Research map / Vision-language and multimodal / Multimodal LLMs
Towards Natural Language-Guided Drones: GeoText-1652 Benchmark with Spatial Relation Matching
Meng Chu, Zhedong Zheng, Wei Ji, Tingyu Wang, Tat-Seng Chua
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 grounding loss (ablation) | XVLM finetuned · GeoText-1652 (text query) | Recall@10: 29.6 -> 30.9 | Improved |
| add spatial relation loss (ablation) | XVLM finetuned · GeoText-1652 (text query) | Recall@10: 29.6 -> 30.1 | Improved |
| combine grounding and spatial losses (ablation) | XVLM finetuned · GeoText-1652 (image query) | Recall@1: 25.0 -> 26.3 | Improved |
| fine-tune on GeoText-1652 instead of zero-shot | XVLM 16M · GeoText-1652 (text query) | Recall@10: 13.4 -> 29.6 | Improved |
| train on satellite + drone + ground instead of drone only (ablation) | Ours · GeoText-1652 (text query) | Recall@1: 12.9 -> 13.6 | Improved |
| train on all three platforms instead of satellite + ground (ablation) | Ours · GeoText-1652 (image query) | Recall@1: 18.7 -> 26.3 | Improved |
| raise spatial loss weight lambda from 0.1 to 1.0 (ablation) | Ours · GeoText-1652 (text query) | Recall@1: 13.6 -> 10.5 | Hurt |
| lower lambda from 0.1 to 0.05 (ablation) | Ours · GeoText-1652 (text query) | Recall@1: 13.6 -> 12.3 | Hurt |
| rotate test images by 90 degrees (ablation) | Ours · GeoText-1652 (text query) | Recall@10: 31.2 -> 29.6 | Hurt |
| increase pretraining images from 4M to 16M in XVLM zero-shot | XVLM · GeoText-1652 (text query) | Recall@10: 13.2 -> 13.4 | No clear effect |
Closest papers by meaning
- Learning Fine-Grained Alignment for Aerial Vision-Dialog Navigation (AAAI 2025 · similarity 0.87)
- SpatialRGPT: Grounded Spatial Reasoning in Vision-Language Models (NeurIPS 2024 · similarity 0.87)
- Stitch and Tell: A Structured Data Augmentation Method for Spatial Understanding (NeurIPS 2025 · similarity 0.87)
- LangMap: A Human-Verified Benchmark for Hierarchical Open-Vocabulary Goal Navigation (NeurIPS 2026 · similarity 0.86)
- ERGeoBench: A Comprehensive Benchmark for Embodied Reasoning and Geo-localization in Multimodal Large Language Models (ICML 2026 · similarity 0.86)
- InternSpatial: A Comprehensive Dataset for Spatial Reasoning in Vision-Language Models (ICLR 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.
