onlyairnopods / geo-mllm Goto Github PK

1. configuration: config.yaml
2. Use multimodal LLM to inference: infer.py
3. Extract specific results(e.g.: latitude and longitude): extract_info.py
4. Use ChatGPT to check and refine the results: check_GPT.py
5. Change prompt method: method.py
6. Calculate distance between prediction and ground truth: calc_dist.py
7. Automatic script: run.py

[
    {
        "image_file": "file_name",
        "gt": {
            "latitude": "value",
            "longitude": "value"
        },
        "model_1": {
            "method_1": {
                "output": " ",
                "latitude": " ",
                "longitude": " ",
                "location": " ",
                "xxx": " "
            }
        },
        "model_2": {
            "method_1": {
                "output": " ",
                "xxx": " "
            }
        },
    },
    ......
]

• image_file: The name of the image file.
• gt: A nested object containing geographical location information.
  • latitude: The latitude value.
  • longitude: The longitude value.
• model_name: A nested object containing model-related information.
  • method: A nested object containing method-related information.
    • output: The output result of the method.
    • xxx: Other relevant information.

View Interactive Map on local browser

Feel free to run a simple demo:

python infer_demo.py --image-path /path/to/image

or infer_demo.ipynb