The Challenge

Duration: 24 hours

Build a creative computer vision application using:

• Focoos platform for model training
• ONNX export format with quantization library
• Arduino Nicla Vision as deployment target
• Zant for flashing your model

Use Case: Any computer vision application - we'll reward your creativity! 🎨

---

Submission Requirements

Submit via Pull Request to this repository

Required Files

• .ino Arduino sketch
• .onnx model file
• model_info.json metadata
• README.md documentation
• anything relevant (web pages, mobileapp, images, examples ...)
• 5 slides presentation deck (.pdf)
  • Use the Google Slides template here: File → Make a Copy → Entire Presentation

Folder structure

team-number-project-name/ (example: 7-cool-object-detection)
├── src/
│   ├── sketch.ino               # Arduino sketch
│   ├── model.onnx               # ONNX model
│   ├── model_info.json          # Model metadata
│   └── ...                      # (optional) Additional relevant files (web pages, mobile app, etc...)
│
├── docs/
│   ├── README.md                # Extended documentation (setup, usage, pipeline, etc.)
│   └── other-guides.md          # (optional) additional guides or docs
│
├── slides/
│   └── presentation.pdf         # Presentation deck
│
└── README.md                    # Main project readme (overview + quick start)

Selection Process:

• ✅ All submissions reviewed
• 🎤 Top 5 teams selected for live demo
• 🏆 3 winning teams announced

🏆 Award Categories 🏆

🌟 Impact Trophy

Most innovative and original application concept

💻 Clean Code Champion

Best code quality, documentation, and software engineering practices

🚀 Technical Trophy

Most technically impressive or groundbreaking implementation

🏆 Winners 🏆

1. MelaNoMore

2. SpaceDebris

3. CineCla & MoodSip

You can check out all the amazing projects here.

Some Food for Thought

As you build your solution, here are some aspects worth considering:

On Data & Models

• How did you choose your dataset? Edge AI has unique constraints that might influence what data works best
• Why did you pick that specific model architecture? Sometimes smaller is better... 😉
• Have you thought about the trade-offs between accuracy and efficiency for your particular use case?

On Performance & Deployment

• Getting your model running on actual hardware is just the beginning - does it run fast enough for what you're trying to do?
• A surveillance system might need different framerates than a periodic quality checker
• What if some of the all computation didn't happen on the device? Pre-processing, post-processing, or complementary algorithms could live elsewhere

On Integration & User Experience

• The Nicla Vision can communicate with other systems - how reliable is your connection?
• Could you visualize your system's output in a way that anyone could understand, not just developers?
• Have you explored what other sensors the Nicla Vision has beyond the camera? IMU, microphone, environmental sensors...

On Going Further

• What if you ran more than one model? The hardware might surprise you with what it can handle
• Think about the full pipeline: what happens before and after your model runs?
• How would someone without technical knowledge interact with your system?

---

🛠️ Resources

• Focoos Platform: Model training & export
• Zant Tool: Model deployment to hardware
• Arduino Nicla Vision: Edge AI hardware platform

See the docs/ folder for tutorials, troubleshooting, and additional resources

---

Good luck, and remember: creativity counts! 🎉