Building Age Classifier (Zero-shot) using GPT-4 & Facade Image in London (FI-London) Dataset

       

⭐ Highlight

🔥 Zero-shot Classifier: Estimate buinding age from facade image without any training!
🔥 Facade Image Dataset: FI-London combining facade images and building's attributs including builidng age and land use.
⚠️ To test our classifier, the paid OpenAI API Key to call GPT-4 Vision is required.

📋 Table of Contents

• Installation
• FI-London
• Demo
• TODOs
• License
• Acknowledge
• Citation

🛠 Installation

Follow these steps to set up your environment and install the required packages for Building Age Classifier.

Step 1: Create a Conda Environment

conda create -n BaClassifier python=3.8
conda activate BaClassifier

Step 2: Install Necessary Packages for Classifier

pip install openai==1.2.0

Step 3: Install Necessary Packages for Evaluation (Optional)

pip install sklearn matplotlib seaborn numpy

Step 4: Install Building Age Classifier and Download FI-London

git clone https://github.com/3Dimaging-ucl/Building_Age_Classifier_with_FI-London.git

🏢 FI-London

FI-London is now avaiable to reproduce our experiment in our paper or do further test. FI-London will be available soon. Please download FI-London here.

For a start, you should unzip facade images.

unzip FI-London.zip -d ./

The structure of FI-London dataset combining facade images with building attributes has shown below. Building attributes are saved in a json file.

FI-London Structure:

├─FI-London
│  ├─Image
│  │  ├─1.jpg
│  │  ├─2.jpg
│  │  ├─...
│  │  └─131.jpg
│  └─Building_Attribute.json

Building Attribute Structure:

{
    "Dataset Information": {
        "Name": "FI-London",
        "Author": "Zichao Zeng",
        "Date": "07-01-2024",
        "Reference": "ZERO-SHOT BUILDING AGE CLASSIFICATION FROM FACADE IMAGE USING GPT-4",
        "URL": "https://zichaozeng.github.io/ba_classifier",
        "License": "CC BY-NC 4.0: https://creativecommons.org/licenses/by-nc/4.0/deed.en"
    },
    "Data": [
        {
            "ID": 1, # Image ID
            "Image": "1.jpg", # File name
            "City": "Brent", # Collect area
            "Longitude": 51.553552, # Collect location, 
            "Latitude": -0.284703, # not the building location
            "Date": "28/12/2023", # Collect date
            "Creator": "Zichao Zeng", # Creator
            "Age Epoch": "1920-1939", # Label of building age
            "Land Use": "Residential (unverified)", # Label of land use
            "Specific Use 1": "NaN", # Label of land specific use
            "Specific Use 2": "NaN"
        },
    ]
}

🕹️ Demo

To demo our zero-shot classifier, we provide three modes (single image test, multiple images test, and FI-London test). Input image(s) path, output path of predicted results, image mode (single/multiple/FI-London), and BaC mode - prompt setting (1/5) are optional. *If BaC mode is 5, classifier will provide top 5 most likely age epochs, which we are not recommend to implement.

Single Image Demo

python BAC_demo.py ./Test/Image/1.jpg {your openai api key} --output ./Test/Prediction/resutl_single.json --img_mode single --BaC_mode 1

Multiple Images Demo:

python BAC_demo.py ./Test/Image/ {your openai api key} --output ./Test/Prediction/resutl_multiple.json --img_mode multiple --BaC_mode 1

FI-London Demo (default):

python BAC_demo.py ./FI-London/Image/ {your openai api key} --output ./Result_FI-London/resutl_FI-London.json --img_mode FI-London --BaC_mode 1

🗓️ TODOs

✅ Age classifier is realsed.
✅ Age claffication demo is realsed.
✅ Paper is published.
✅ FI-London v1 is released.
🔲 Extend data size of FI-London.
🔲 Land use classifier is on the way.

📐 License

This project is released under MIT licence terms. All data in FI-London is released under CC BY-NC 4.0. Codes of zero-shot classifier and FI-London are free to download, use and share.

❤️ Acknowledge

This zero-shot classifier is based on GPT-4 Vision by OpenAI.

The Facade Image in London (FI-London) dataset is grounded in Colouring Cities Research Programme by the Alan Turing Institude and Colouring London by Hudson, P., et al. in 2018.

We thank for their great works.

📜 Citation

Should our work offer you even the slightest inspiration, we would be most honoured if you chose to cite our paper.

@article{zeng2024zeroshot,
        title={Zero-shot Building Age Classification from Facade Image Using GPT-4},
        author={Zichao Zeng and June Moh Goo and Xinglei Wang and Bin Chi and Meihui Wang and Jan Boehm},
        journal={arXiv preprint arXiv:2404.09921},
        year={2024}
}