# Aerial Object Detection: Rural Buildings

### 🏆 Challenge Summary  
Welcome to **DualityAI and LunateAI's Geospatial Object Detection** challenge!  In this challenge, participants compete to get the best real-world object detection results while training their model with **100% synthetic data**. This competition is based on **real-world AI applications** where models trained in controlled virtual environments must perform well in complex or more challenging conditions. In this instance, the real world-data contains smaller, lower resolution, and/or less contrasted target detections.

Participants will train a model on **synthetic aerial images** of a rural landscape, generated by by participants in  [FalconCloud](https://falcon.duality.ai/secure/scenarios/edit/0b96f95d-adff-49f5-82a3-9fbffebcb7f1?utm_source=kaggle&utm_medium=kaggle_competition&utm_campaign=competition_5), Duality AI’s digital twin simulation software. Then, users will test the model’s performance on previously **unseen real-world images**. 

**What's In it for you?**  
- **Win prizes**: Money awards*, Certificates and Shoutouts on platforms like LinkedIn and Hugging Face with a chance to feature on our competition blog.  
       - **[*See the list of countries that CAN receive the prize money](https://docs.google.com/spreadsheets/d/1gGBsSrGXpkfbpNhUe3CzqFNOHwFYVF0kKeEh3npt6m0/edit?usp=sharing)**. All other countries have restricted money transfer from the US.
       - $250 for first place, $100 for second place, $50 for third place
- **Master synthetic data on the cloud**: Learn skills applicable to real-world AI deployment.  
- **Boost your portfolio**: Showcase your AI & ML skills in real-world object detection.
- **Stand out to recruiters**: Rank high on Kaggle, build credibility, and showcase your skills on LinkedIn, GitHub, and resumes.
- **[Network & grow](https://discord.com/invite/dualityfalconcommunity)**: Compete with fellow students, ML engineers, researchers, collaborating on our Discord to answer questions and build connections.

So, how well can your model **detect objects in real-world images after training only on synthetic data**? Let’s find out!   

---

### 🎯 **Objective**
- **Train on synthetic data**: Generate your own datasets on FalconCloud and decide which parameters produce the highest-quality data for optimal model performance.
- **Test on real-world images**: Detect buildings in real-world aerial images taken from a drone or satellite. 
- **Score using mAP@50**: Achieve the highest **mAP@50** (Mean Average Precision at IoU 0.5) score.   

---

### 💡 **Why Synthetic Data?**
Synthetic data is **cost-effective, scalable, and bias-free** – yet synthetic-trained models often fail in real-world conditions  (referred to as the Sim2Real gap). Since 2018 [Duality](https://www.duality.ai/?utm_source=kaggle&utm_medium=kaggle_competition&utm_campaign=competition_5) has been building tools and strategies that bridge this gap, and we’re sharing them with you! 

For this competition, Duality is partnering with [LunateAI](https://lunateai.com/) - a high-end advisory business founded by the award-winning, industry-recognized global leader Dr. Nadine Alameh to usher in a new era of geospatial impact in conjunction with advances in computing and AI. 

The skills you develop in this challenge apply to: 

- Autonomous robots 
- Industrial automation  
- AI-driven quality control  
- Augmented reality (AR) & Computer Vision applications
- And many more cutting-edge technologies!

**Featured Tool**: [Falcon](http://duality.ai/edu?utm_source=kaggle&utm_medium=kaggle_competition&utm_campaign=competition_5)  
- Free subscription tier available, Sign up [here](https://falcon.duality.ai/secure/scenarios/edit/0b96f95d-adff-49f5-82a3-9fbffebcb7f1?utm_source=kaggle&utm_medium=kaggle_competition&utm_campaign=competition_5)   
- Generates 300+ high-quality synthetic images in minutes  
- Customize time of day, land features, and more!  
- See example FalconCloud scenarios [here on our user site](https://falcon.duality.ai/secure/scenarios?utm_source=kaggle&utm_medium=kaggle_competition&utm_campaign=competition_5)