2024 EY Open Science Data Challenge: Tropical Storm Damage Detection Model

Team: Double Y (Yi Jie Wong, Yin Loon Khor, Liu Ziwei)

The goal of the challenge is to develop a machine learning model to identify and detect “damaged” and “un-damaged” coastal infrastructure (residential and commercial buildings), which have been impacted by natural calamities such as hurricanes, cyclones, etc. Participants will be given pre- and post-cyclone satellite images of a site impacted by Hurricane Maria in 2017 and build a machine learning model, designed to detect four different objects in a satellite image of a cyclone impacted area:

1. Undamaged residential building
2. Damaged residential building
3. Undamaged commercial building
4. Damaged commercial building

Our Solution

• We are among the Top 10 Global Semi-Finalist of EY Open Science Data Challenge 2024 🎉🥳
  
• We ranked 8th in Phase 1 out of 11,000 registrants! 🌍🏆
  
• In terms of evaluation score, we rank 4th, tying with other impressive competing teams! 🤩
• Meanwhile, we ranked 1st out of 22 teams in Malaysia! 🏅

Repo Structure

EY-challenge-2024
├── our-best-runs                       (proof of our experiment that yields the highest mAP)
│   ├── detect
│   │   ├── predict                     
│   │   ├── train                       
├── additional-dataset.zip              (additional dataset)
├── best-trained-model.pt               (best trained model which we used for submission, mAP 0.51)
├── challenge_1_submission_images.zip   (just the zip file of EY Challenge Phase 1 test images)
├── labelled-dataset.zip                (labelled dataset)
├── Model-development-notebook.ipynb    (to train the model)
├── pretrained-on-msft-puerto-rico      (models pretrained on Microsoft Building Footprint -> Puerto Rico dataset)
├── requirements.txt                    (dependencies requirement)
├── Validation-notebook.ipynb           (for Phase 1 submission)

Setup environment

# Clone the repo
git clone https://github.com/yjwong1999/EY-challenge-2024.git
cd EY-challenge-2024

# Create conda environment
conda create --name ey-challenge python=3.8.10 -y
conda activate ey-challenge

Training

1. Use Model-development-notebook.ipynb to train the model following our pipeline. If you cant download the additional-dataset.zip due to filesize error, than you can use Model-development-notebook (backup).ipynb as an alternative
2. Use Validation-notebook.ipynb to generate submission.zip

Download dataset

1. Our submitted content-package can be found in this link. You can find all of our datasets here as well

Cite this repository

@article{Wong2024,
   author = {Yi Jie Wong and Yin Loon Khor and Liu Ziwei},
   doi = {10.36227/TECHRXIV.172963135.56918790/V1},
   institution = {Techrxiv},
   journal = {Authorea Preprints},
   month = {10},
   publisher = {Authorea},
   title = {Automating Coastal Vulnerability Assessment: AI-Driven Geospatial Analysis via Building Damage Detection},
   url = {https://www.authorea.com/users/844381/articles/1233489-automating-coastal-vulnerability-assessment-ai-driven-geospatial-analysis-via-building-damage-detection},
   year = {2024},
}