A beginner's guide to the Yolo11n model by Ultralytics on Replicate

#coding #ai #machinelearning #programming

This is a simplified guide to an AI model called Yolo11n maintained by Ultralytics. If you like these kinds of analysis, you should join AImodels.fyi or follow us on Twitter.

The yolo11n represents the nano variant of Ultralytics' latest YOLO11 object detection architecture, designed for maximum efficiency with minimal computational overhead. This compact model contains just 2.6M parameters while achieving 39.5 mAP50-95 on the COCO dataset, making it the smallest and fastest option in the YOLO11 family. Compared to previous generations like YOLOv8, the YOLO11 series introduces architectural improvements that boost both accuracy and inference speed. Built by Ultralytics, this model excels in real-time applications where computational resources are constrained, processing images in just 1.55 milliseconds on T4 GPU hardware.

Model inputs and outputs

The model accepts standard image inputs and provides comprehensive object detection results with flexible configuration options. Users can fine-tune detection sensitivity through confidence and IoU thresholds, while supporting various image sizes to balance speed and accuracy based on application requirements.

Inputs

image: Input image in URI format for object detection processing
conf: Confidence threshold (0-1, default 0.25) to filter low-confidence detections
iou: IoU threshold (0-1, default 0.45) for non-maximum suppression to eliminate duplicate detections
imgsz: Image size selection (320, 416, 512, 640, 832, 1024, 1280 pixels, default 640) affecting processing speed and detection quality

Outputs

image: Processed image with bounding boxes and labels drawn around detected objects
json_str: Structured JSON containing detection results including coordinates, confidence scores, and class labels