Research map / Vision-language and multimodal / Multimodal LLMs
EarthVL: A Progressive Earth Vision-Language Understanding and Generation Framework
Junjue Wang, Yanfei Zhong, Zihang Chen, Zhuo Zheng, Ailong Ma, Liangpei 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 segmentation guidance (ablation) | EarthVLNet · EarthVLSet multiple-choice VQA | OA (%): 79.63 -> 81.06 | Improved |
| add segmentation guidance (ablation) | EarthVLNet · EarthVLSet multiple-choice VQA | RMSE: 2.636 -> 2.340 | Improved |
| add segmentation guidance (ablation) | EarthVLNet · EarthVLSet open-ended VQA | LAVE (%): 77.94 -> 80.44 | Improved |
| add segmentation guidance (ablation) | EarthVLNet · EarthVLSet open-ended VQA | BLEU1: 0.5653 -> 0.5726 | Improved |
| add object-guided attention (OGA) instead of SE channel attention (ablation) | EarthVLNet · EarthVLSet multiple-choice VQA | OA (%): 80.72 -> 81.06 | Improved |
| add object-guided attention instead of spatial attention SA (ablation) | EarthVLNet · EarthVLSet multiple-choice VQA | OA (%): 79.83 -> 81.06 | Improved |
| add spatial attention SA on concatenated features (ablation) | EarthVLNet · EarthVLSet multiple-choice VQA | OA (%): 80.21 -> 79.83 | Hurt |
| add object guidance with OGA vs only features (ablation) | EarthVLNet · EarthVLSet multiple-choice VQA | OA (%) / OR: 79.97 -> 81.06; 2.582 -> 2.340 | Improved |
| use ND-Separated optimization instead of CE loss (ablation) | EarthVLNet · EarthVLSet multiple-choice VQA | OA (%): 79.91 -> 81.06 | Improved |
| use ND-Separated instead of ND-Shared (ablation) | EarthVLNet · EarthVLSet multiple-choice VQA | OA (%): 80.63 -> 81.06 | Improved |
| use DIW loss instead of CE loss (ablation) | EarthVLNet · EarthVLSet multiple-choice VQA | OA (%): 79.91 -> 79.51 | Hurt |
| scale up ConvNeXt vision encoder (ablation) | Semantic-FPN + FlanT5-XL · EarthVLSet open-ended VQA | BLEU1: 0.505 -> 0.564 | 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.
- GeoChat — GeoChat: Grounded Large Vision-Language Model for Remote Sensing (CVPR 2024) · 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
- EarthVQA: Towards Queryable Earth via Relational Reasoning-Based Remote Sensing Visual Question Answering (AAAI 2024 · similarity 0.94)
- VRSBench: A Versatile Vision-Language Benchmark Dataset for Remote Sensing Image Understanding (NeurIPS 2024 · similarity 0.90)
- VLRS-Bench: A Vision-Language Reasoning Benchmark for Remote Sensing (NeurIPS 2026 · similarity 0.90)
- LISAt: Language-Instructed Segmentation Assistant for Satellite Imagery (NeurIPS 2025 · similarity 0.90)
- Towards Faithful Reasoning in Remote Sensing: A Perceptually-Grounded GeoSpatial Chain-of-Thought for Vision-Language Models (ICLR 2026 · similarity 0.90)
- TEOChat: A Large Vision-Language Assistant for Temporal Earth Observation Data (ICLR 2025 · similarity 0.89)
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.
