Module 9: Advanced Deep LearningBy Ashok Kumar Yadav4 min read

YOLO Object Detection Tutorial in Python

YOLO (You Only Look Once) predicts boxes and classes in a single forward pass, which is why it is the default choice for real-time detection. You will install it, run detection, and train on your own labels.

How YOLO detects objects

Older detectors proposed regions first and classified them second. YOLO treats detection as one regression problem: the network outputs bounding boxes, confidence and class scores directly. Non-maximum suppression then removes overlapping duplicates, so only the best box per object remains.

Install and run detection

The Ultralytics package bundles models, training and export. The nano model is small enough for CPUs and a good first test.

# pip install ultralytics
from ultralytics import YOLO

model = YOLO("yolov8n.pt")          # downloads weights on first run
results = model("street.jpg")       # list of Results objects
r = results[0]
for box in r.boxes:
    cls = r.names[int(box.cls)]
    print(cls, round(float(box.conf), 2), box.xyxy[0].tolist())
r.save("street_detected.jpg")

Real-time detection from a webcam

Stream frames with OpenCV and let YOLO draw the boxes. Press q to quit.

import cv2
from ultralytics import YOLO

model = YOLO("yolov8n.pt")
cap = cv2.VideoCapture(0)
while cap.isOpened():
    ok, frame = cap.read()
    if not ok:
        break
    annotated = model(frame, conf=0.4, verbose=False)[0].plot()
    cv2.imshow("YOLO", annotated)
    if cv2.waitKey(1) & 0xFF == ord("q"):
        break
cap.release()
cv2.destroyAllWindows()

Train YOLO on a custom dataset

Label images in YOLO format (one .txt file per image with class x_center y_center width height, all normalised). Then describe the dataset in a YAML file and fine-tune from pretrained weights.

# data.yaml
# path: datasets/helmets
# train: images/train
# val: images/val
# names:
#   0: helmet
#   1: no_helmet

from ultralytics import YOLO

model = YOLO("yolov8n.pt")
model.train(data="data.yaml", epochs=50, imgsz=640, batch=16)
metrics = model.val()
print("mAP50-95:", metrics.box.map)
model.export(format="onnx")

Fixing common problems

Low recall usually means too few labelled examples or a confidence threshold that is too high. Many false positives suggest background images are missing from training. Always compare train and validation loss: if validation worsens while training improves, add data or augmentation instead of training longer.

from ultralytics import YOLO
model=YOLO('yolo11n.pt')
r=model('image.jpg',verbose=False)[0]
print(len(r.boxes))
print(r.boxes.conf[:3].cpu().numpy())

<number of detections>
[confidence values]

def iou(a,b):
    x1,y1=max(a[0],b[0]),max(a[1],b[1]); x2,y2=min(a[2],b[2]),min(a[3],b[3])
    inter=max(0,x2-x1)*max(0,y2-y1)
    aa=(a[2]-a[0])*(a[3]-a[1]); bb=(b[2]-b[0])*(b[3]-b[1])
    return inter/(aa+bb-inter)
print(round(iou([0,0,10,10],[5,5,15,15]),3))

0.143

Frequently asked questions

Primary references

Use these official sources when exact syntax, defaults or definitions matter beyond the examples here.

• Ultralytics YOLO documentation
• OpenCV Python tutorials
• You Only Look Once (Redmon et al.)