Military Object Detection — YOLOv8n

A fine-tuned YOLOv8 nano model for detecting military and civilian objects in images.
Trained on a custom military imagery dataset covering 12 object categories.

---

Model Description

Property	Value
Architecture	YOLOv8n (nano)
Parameters	~3.0 M
GFLOPs	8.2
Model size	24.5 MB
Task	Object Detection
Input size	640 × 640
Framework	Ultralytics 8.x

---

Dataset

A custom-collected military imagery dataset containing annotated images of battlefield and civilian scenes.

Property	Value
Number of classes	12
Annotation format	YOLO (normalized bounding boxes)
Image sources	Open-source military imagery
Augmentations	Mosaic, flip, HSV shift, scale

Class Names

ID	Class
0	camouflage_soldier
1	weapon
2	military_tank
3	military_truck
4	military_vehicle
5	civilian
6	soldier
7	civilian_vehicle
8	military_artillery
9	trench
10	military_aircraft
11	military_warship

---

Training Configuration

Hyperparameter	Value
Base model	YOLOv8n
Optimizer	AdamW (auto)
Epochs	100
Image size	640
Batch size	16
Confidence threshold (inference)	0.40
IoU threshold (NMS)	0.50
Device	CPU / CUDA

---

Performance Metrics

Metrics measured on the held-out validation split.

Metric	Value
mAP@50	~0.72
mAP@50-95	~0.48
Precision	~0.74
Recall	~0.68
Inference speed (CPU, 320 px)	~120 ms/image

Note: Exact per-class metrics depend on dataset split and augmentation seed.

---

Inference

Install dependencies

pip install ultralytics

Load from Hugging Face Hub

from huggingface_hub import hf_hub_download
from ultralytics import YOLO

# Download weights
model_path = hf_hub_download(
    repo_id="datasidahmed/YOLOV8",
    filename="best.pt"
)

# Load model
model = YOLO(model_path)

Or load directly by filename

from ultralytics import YOLO

model = YOLO("best.pt")   # if best.pt is already in the working directory

Run inference

from huggingface_hub import hf_hub_download
from ultralytics import YOLO

model_path = hf_hub_download(repo_id="datasidahmed/YOLOV8", filename="best.pt")
model = YOLO(model_path)

# Single image
results = model.predict("image.jpg", conf=0.40, iou=0.50)

# Display results
for r in results:
    for box in r.boxes:
        cls_id   = int(box.cls[0])
        conf     = float(box.conf[0])
        x1,y1,x2,y2 = map(int, box.xyxy[0])
        print(f"{model.names[cls_id]}: {conf:.2f}  [{x1},{y1},{x2},{y2}]")

# Save annotated image
results[0].save("output.jpg")

Batch inference on a folder

results = model.predict("images/", conf=0.40, save=True)

Export to ONNX

model.export(format="onnx", imgsz=640)

---

Limitations

• Domain specificity — trained on a specific military imagery corpus; performance may degrade on imagery with uncommon lighting, extreme viewpoints, or non-standard camouflage patterns.
• Small-object detection — as a nano (n) variant, the model trades accuracy for speed; larger variants (YOLOv8s/m/l) may perform better on distant or small targets.
• Class imbalance — rare classes such as military_warship, military_aircraft, and trench have fewer training samples and may exhibit lower recall.
• Ethical use — this model is intended for research, simulation, and defensive awareness applications. Use in live operational systems requires additional validation and appropriate human oversight.
• Not a weapons system — detections are bounding-box predictions with confidence scores. They must not be used as the sole basis for any consequential decision.

---

Citation

If you use this model in your research or project, please cite:

@misc{melainin2024militarydetection,
  author    = {Sidahmed Melainin},
  title     = {Military Object Detection using YOLOv8},
  year      = {2024},
  publisher = {Hugging Face},
  url       = {https://huggingface.co/datasidahmed/YOLOV8}
}

---

Author

Sidahmed Melainin
GitHub: Melainin2