Charting New Territories

Charting New Territories: Exploring the Geographic and Geospatial Capabilities of Multimodal LLMs

Jonathan Roberts, Timo Lüddecke, Rehan Sheikh, Kai Han, Samuel Albanie

We conduct a series of experiments exploring various vision capabilities of multimodal large language models (MLLMs) within these domains, particularly focusing on the frontier model GPT-4V, and benchmark its performance against open-source counterparts. Our methodology involves challenging these models with a small-scale geographic benchmark consisting of a suite of visual tasks, testing their abilities across a spectrum of complexity. The analysis uncovers not only where such models excel, including instances where they outperform humans, but also where they falter, providing a balanced view of their capabilities in the geographic domain.

Repo Overview

• Experiments Taxonomy
• Takeaways
• Data
• Prompts
• Citation
• Questions

Experiments Taxonomy

• Localisation
  • GeoGuessr
• Remote sensing
  • Classification
  • Change detection
  • Segmentation
  • Bounding boxes
  • Counting
• Mapping
  • Region identification
    • State name from outline
    • City name from maps
    • Island and water body naming from maps
  • Localisation
    • Real-world -> map
    • Map -> real-world
• Flags
  • Identification
• Failure cases
  • Identifying multiple states

Key takeaways

• Of all the evaluated models, GPT-4V can perform the broadest range of tasks. However, it does not always perform best, e.g., satellite image detection and classification tasks. In general, it recognizes fine-detail well but tends to fail when precise localisation is required.
• More broadly, the best model choice depends on the task at hand. Qwen-VL and LLaVA-1.5 in particular often demonstrate good localisation performance.
• Enforcing a specific output format is challenging, models often resort to explanations why they are not capable of performing the task. Among the evaluated models GPT-4V was least susceptible to this behaviour.
• The current generation of leading MLLMs suffer a performance penalty when processing multi-object images, relative to their performance on single object images.

Data

Data for the majority of the experiments can be found in Data.

Prompts

Coming soon!

Citation

If you found our work useful in your own research, please consider citing our paper:

@misc{roberts2023charting,
      title={Charting New Territories: Exploring the Geographic and Geospatial Capabilities of Multimodal LLMs}, 
      author={Jonathan Roberts and Timo Lüddecke and Rehan Sheikh and Kai Han and Samuel Albanie},
      year={2023},
      eprint={2311.14656},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Questions

If you have any questions about our work, please open an issue in this repository.