# Multi-Class Object Detection Challenge 🚀

## 🏆 Challenge Summary

Welcome to the continuation of [Duality AI](https://www.duality.ai)'s Synthetic-to-Real Object Detection Challenges - now with more classes!

Check out some of our previous (finished) competitions to see what we've been building:
- [Multi-class Object Detection](https://www.kaggle.com/competitions/multi-class-object-detection-challenge)
- [Multi-instance Object Detection](https://www.kaggle.com/competitions/multi-instance-object-detection-challenge)
- [Synthetic2Real Object Detection Challenge 2](https://www.kaggle.com/competitions/synthetic-2-real-object-detection-challenge-2)
- [Synthetic to Real Object Detection Challenge](https://www.kaggle.com/competitions/synthetic-2-real-object-detection-challenge)

###Your Goal for this competition: 
Build an object detection model that can identify **multiple classes** (soup + cheerios + candles), with variety within the classes (multiple types of candles). Train on synthetic data from the free FalconCloud simulation (the cloud version of Duality’s high-fidelity digital twin simulation platform), and test on **real-world test images**. 

See how high quality digital twins with carefully crafted simulation can bridge the sim2real gap!

---

###  Who is this competition for?

**Learners wanting to develop skills in:**
- Vision systems
- Object detection
- Data improvement/analysis
- Synthetic data and simulation

---

### 🌟 What’s New in This Challenge?

- **Three object classes**: Cheerios + Soup + Candle
- [**Generate Data in FalconCloud**](https://falcon.duality.ai/secure/scenarios/edit/7c5b51d0-553c-4bd3-9566-73723bbc43a5?utm_source=kaggle&utm_medium=kaggle_competition&utm_campaign=competition_fk1): Generate data on the cloud, train, and tweak parameters to create better data for higher performance!

---

### 🧠 Learn the Real-World Workflow

In this challenge, you’ll follow a simulation-to-deployment pipeline used by actual AI engineering teams:

1. **Generate synthetic training data** using FalconCloud  
2. **Train an object detection model** (YOLO recommended)  
3. **Analyze model performance and Identify weaknesses** in your model such as persistent false positives or false negatives
4. **Collect more data** from the Falcon scenario to address those gaps  
5. **Repeat** the process to achieve the highest mAP@50


This mimics the workflow used in fields like robotics, manufacturing QA, and autonomous systems - where collecting real-world data is expensive and iterative simulation is key.

---

### 💡 Why This Challenge Matters

Synthetic data is **scalable, bias-controllable, and cost-effective**, but it still requires domain alignment to perform in the real world. **Duality** has spent years building technology to bridge that **Sim2Real gap**, and now you get hands-on access to those tools - for free.

---

Ready to build an object detector that adapts, learns, and generalizes - all from simulation? Let’s get started!