YOLO Quick Start: Model Loading and Inference

Fri, 08 May 2026 00:00:00 +0000

Model Loading and Inference Across Versions

Ultralytics Unified API (Works with v8/11/26)

python


 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22


from ultralytics import YOLO

# ========== YOLOv8 ==========
model_v8 = YOLO("yolov8n.pt") # nano
model_v8 = YOLO("yolov8s.pt") # small
model_v8 = YOLO("yolov8m.pt") # medium
model_v8 = YOLO("yolov8l.pt") # large
model_v8 = YOLO("yolov8x.pt") # extra large

# ========== YOLO11 ==========
model_11 = YOLO("yolo11n.pt") # nano
model_11 = YOLO("yolo11s.pt") # small
model_11 = YOLO("yolo11m.pt") # medium
model_11 = YOLO("yolo11l.pt") # large
model_11 = YOLO("yolo11x.pt") # extra large

# ========== YOLO26 (2026 latest) ==========
model_26 = YOLO("yolo26n.pt") # nano recommended for edge deployment
model_26 = YOLO("yolo26s.pt") # small
model_26 = YOLO("yolo26m.pt") # medium
model_26 = YOLO("yolo26l.pt") # large
model_26 = YOLO("yolo26x.pt") # extra large

Image Detection Hands-on Example

python


 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22


from ultralytics import YOLO

# Load model (YOLO26 example)
model = YOLO("yolo26n.pt")

# Single image detection
results = model("test.jpg", conf=0.25, iou=0.45)

# Process results
for result in results:
 boxes = result.boxes # Detection boxes
 masks = result.masks # Segmentation masks
 probs = result.probs # Classification probabilities
 
 # Print detection results
 for box in boxes:
 print(f"Class: {result.names[int(box.cls)]}, "
 f"Confidence: {box.conf.item():.3f}, "
 f"Coordinates: {box.xyxy.tolist()[0]}")
 
 # Save visualization results
 result.save("result.jpg")

Video Detection Hands-on Example

python


 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17


from ultralytics import YOLO

model = YOLO("yolo26n.pt")

# Video file detection
results = model.predict(
 source="input.mp4",
 save=True, # Save result video
 conf=0.3,
 show=False, # Whether to display in real-time
 stream=True # Stream processing to save memory
)

# Process frame by frame
for result in results:
 # Custom post-processing logic
 pass

Real-time Camera Detection

python


 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26


from ultralytics import YOLO
import cv2

model = YOLO("yolo26n.pt")

# Open camera
cap = cv2.VideoCapture(0) # 0 is default camera

while cap.isOpened():
 ret, frame = cap.read()
 if not ret:
 break
 
 # Inference
 results = model(frame, verbose=False)
 
 # Draw results
 annotated_frame = results[0].plot()
 
 # Display
 cv2.imshow("YOLO Real-time", annotated_frame)
 if cv2.waitKey(1) & 0xFF == ord('q'):
 break

cap.release()
cv2.destroyAllWindows()

Version-specific Code Differences

Feature	YOLOv8	YOLO11	YOLO26	YOLOv9	YOLOv10
Unified API	✅	✅	✅	❌ Separate repo	❌ Separate repo
No NMS	❌	❌	✅	❌	✅
DFL Module	✅	✅	❌ Removed	✅	✅
MuSGD Optimizer	❌	❌	✅	❌	❌
Export Compatibility	Good	Good	Best	Fair	Fair

Results Object API Deep Dive

The model() or model.predict() call returns a list of Results objects. Each Results object encapsulates all inference outputs for a single image. Understanding its internal structure is essential for downstream processing.

Inference on Mi&Bee Blog