Yolov8n Proctoring

🎓 YOLOv8n — GradicAI Exam Proctoring

YOLOv8 Nano model used for real-time AI-powered exam proctoring in the GradicAI platform.

Model Description

This is the standard YOLOv8n (nano) model from Ultralytics, pre-trained on the COCO dataset. It is used as-is (no fine-tuning) for the GradicAI proctoring service to detect prohibited objects and suspicious behavior during online exams.

Property	Value
Architecture	YOLOv8 Nano
Parameters	3.2M
Model Size	~6 MB
Training Data	COCO (80 classes)
Input Resolution	640×640
Inference Speed	~100-300ms (CPU)

Classes Used for Proctoring

Out of the 80 COCO classes, GradicAI uses only 3 for proctoring:

COCO Class ID	Label	Proctoring Action
0	person	Count persons — warn if 0 or >1
67	cell phone	⚠️ Warning: phone detected
73	book	⚠️ Warning: study material detected

Violation Logic

if person_count == 0  → "face_absent" warning
if person_count > 1   → "multiple_persons" warning
if phone detected     → "phone" warning
if book detected      → "book" warning

3 warnings → exam terminated

How to Use

Installation

pip install ultralytics

Python Inference

from ultralytics import YOLO

# Load model
model = YOLO("yolov8n.pt")

# Run inference on an image
results = model("exam_frame.jpg", conf=0.25)

# Process detections
for box in results[0].boxes:
    cls = int(box.cls[0])
    conf = float(box.conf[0])
    label = results[0].names[cls]
    print(f"Detected: {label} ({conf:.2f})")

Integration with GradicAI Proctoring

import base64
import io
import numpy as np
from PIL import Image
from ultralytics import YOLO

PERSON_CLASS = 0
PHONE_CLASS = 67
BOOK_CLASS = 73

model = YOLO("yolov8n.pt")

def analyze_frame(b64_data: str) -> dict:
    """Analyze a base64-encoded webcam frame for proctoring violations."""
    if "," in b64_data:
        b64_data = b64_data.split(",", 1)[1]
    
    raw = base64.b64decode(b64_data)
    img = Image.open(io.BytesIO(raw)).convert("RGB")
    frame = np.array(img)
    
    results = model(frame, verbose=False, conf=0.25)[0]
    
    persons = 0
    violations = []
    
    for box in results.boxes:
        cls = int(box.cls[0])
        if cls == PERSON_CLASS:
            persons += 1
        elif cls == PHONE_CLASS:
            violations.append("phone")
        elif cls == BOOK_CLASS:
            violations.append("book")
    
    if persons == 0:
        violations.append("face_absent")
    elif persons > 1:
        violations.append("multiple_persons")
    
    return {"violations": violations, "person_count": persons}

Part of GradicAI

This model powers the proctoring feature of GradicAI — an AI-powered exam and grading platform:

• 🤖 AI Grading — GPT-4o grades open-ended answers with per-question feedback
• 📹 Live Proctoring — YOLOv8n webcam monitoring via WebSocket
• 📝 Structured Exams — Google Forms-style per-question UI
• 📚 AI Quizzes — Auto-generated practice quizzes from exam content

License

The YOLOv8 model is released under the AGPL-3.0 license by Ultralytics.

Citation

@software{yolov8_ultralytics,
  author = {Glenn Jocher and Ayush Chaurasia and Jing Qiu},
  title = {Ultralytics YOLOv8},
  version = {8.0.0},
  year = {2023},
  url = {https://github.com/ultralytics/ultralytics},
  license = {AGPL-3.0}
}