YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
Weapon Detection YOLOv8n v2
Fine-tuned YOLOv8n for weapon detection in CCTV surveillance.
Model Details
- Base: Subh775/Threat-Detection-YOLOv8n
- Architecture: YOLOv8n (3M params)
- Input size: 960px (critical for accuracy — see improvements below)
- Training: 480 images, 50 epochs, fine-tuned 2026-04-07
Classes
| ID | Name |
|---|---|
| 0 | Gun (handgun, pistol, rifle, shotgun) |
| 1 | explosion |
| 2 | grenade |
| 3 | knife (blade, sword, dagger) |
Improvements over v1 (imgsz=960)
| Test Image | v1@640 | v2@960 |
|---|---|---|
| gun_holding_03 | Gun(0.10) | Gun(0.89) |
| gun_fullbody_04 | Gun(0.12) | Gun(0.77) |
| knife_hoodie_01 | NONE | knife(0.80) |
| knife_woman_04 | Gun(0.07) wrong | knife(0.60) correct |
Usage
from ultralytics import YOLO
model = YOLO("weights/best.pt")
results = model("image.jpg", imgsz=960, conf=0.4)
TRT Deployment
weights/best_960.engine — pre-exported for imgsz=960 FP16.
Requires classes.json sidecar for TRT class name mapping.
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support