Update README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,54 @@
|
|
| 1 |
-
---
|
| 2 |
-
license: mit
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
language:
|
| 4 |
+
- en
|
| 5 |
+
metrics:
|
| 6 |
+
- precision
|
| 7 |
+
- recall
|
| 8 |
+
- accuracy
|
| 9 |
+
pipeline_tag: object-detection
|
| 10 |
+
tags:
|
| 11 |
+
- yolo26n
|
| 12 |
+
- onnx
|
| 13 |
+
---
|
| 14 |
+
🔥 Fire & Smoke Detection — YOLO‑26n (ONNX)
|
| 15 |
+
Model Card
|
| 16 |
+
📝 Overview
|
| 17 |
+
This model is a fire and smoke detection system built using YOLO‑26n (YOLO‑NAS) and exported to ONNX format for fast, lightweight deployment. It is trained for 100 epochs on a custom dataset containing annotated fire and smoke images. The model is optimized for CPU‑only environments, making it suitable for real‑time safety monitoring on edge devices, CCTV systems, and low‑resource hardware.
|
| 18 |
+
The detector identifies two critical hazard classes:
|
| 19 |
+
- Fire (Flames)
|
| 20 |
+
- Smoke
|
| 21 |
+
By analyzing each frame of a video stream, the model predicts bounding boxes, confidence scores, and class labels, enabling early detection of hazardous conditions before traditional sensors react.
|
| 22 |
+
|
| 23 |
+
📊 Performance Metrics (100 Epochs)
|
| 24 |
+
|
| 25 |
+
- AP50 (74.3%) — Strong detection accuracy for flames and smoke at IoU ≥ 0.50.
|
| 26 |
+
- mAP50‑95 (43.9%) — Good robustness across stricter IoU thresholds, especially for smoke.
|
| 27 |
+
- Precision (73.6%) — Low false‑positive rate; reliable for real‑world alerts.
|
| 28 |
+
- Recall (68.3%) — Detects most real fire/smoke events, reducing missed hazards.
|
| 29 |
+
These metrics indicate a balanced, reliable model suitable for early hazard detection.
|
| 30 |
+
|
| 31 |
+
⚙️ Model Details
|
| 32 |
+
- Architecture: YOLO‑26n (YOLO‑NAS)
|
| 33 |
+
- Format: ONNX
|
| 34 |
+
- Input Size: 640 × 640
|
| 35 |
+
- Batch Size: 1
|
| 36 |
+
- Precision: FP32 (CPU‑friendly)
|
| 37 |
+
- Post‑processing: Custom NMS + confidence filtering
|
| 38 |
+
- Training Epochs: 100
|
| 39 |
+
|
| 40 |
+
🌍 Use Cases
|
| 41 |
+
- Forest fire early‑warning systems
|
| 42 |
+
- Industrial safety monitoring
|
| 43 |
+
- Warehouse and factory surveillance
|
| 44 |
+
- Smart building fire detection
|
| 45 |
+
- CCTV‑based hazard detection
|
| 46 |
+
- Drone‑based fire inspection
|
| 47 |
+
|
| 48 |
+
📦 Intended Audience
|
| 49 |
+
- Researchers working on fire‑safety automation
|
| 50 |
+
- Developers building real‑time hazard detection systems
|
| 51 |
+
- Smart‑city and industrial monitoring teams
|
| 52 |
+
- Students and engineers exploring ONNX deployment
|
| 53 |
+
|
| 54 |
+
Author: Darshan Modi
|