The EdgeFire Challenge 2025 is a beginner-level AI competition focused on real-time forest fire detection using lightweight, edge-deployable machine learning models. Participants will design, train, and evaluate models capable of identifying forest fires from visual or sensor-based data while meeting the constraints of low-power hardware like Raspberry Pi, Jetson Nano, and similar edge devices.

Your task is to create a novel yet efficient model that can:

Accurately detect fire or smoke in real-time images or sensor readings.

Run smoothly on limited hardware without relying heavily on cloud computation.

Maintain a small model size and minimal inference latency.

You may use any open-source framework (TensorFlow Lite, PyTorch Mobile, ONNX, etc.) and can apply optimization techniques such as:

Quantization (to reduce precision and improve inference speed)

Pruning (to remove redundant parameters)

Knowledge distillation (to create smaller student models from large ones)

Custom architecture design (for creativity and novelty)

A Forest Fire Detection Dataset is provided. However, participants are free to:

Download and preprocess the dataset as they see fit.

Shuffle, split, or augment the data for better performance.

Create and publish a modified public version (with credits and documentation).

Clearly specify all dataset preprocessing steps in their submitted notebook/report under a section titled “Data Preparation.”

This competition aims to inspire new developers and students to explore AI for environmental protection, blending innovation, technical skill, and sustainability.