🟠 YOLOv5 Detection

Function Description

The operator is based on deep learning models of YOLO architecture, capable of identifying multiple targets in images and outputting their class, confidence score and position information for each target. Supports uploading both .onnx and .epicnn file formats.

Usage Scenarios

Suitable for quickly finding object positions in images (without providing object rotation angles), fast positioning, tracking targets and counting quantities. For smart cameras, efficient detection can be achieved by uploading .epicnn models.

Input Output

Input

Image: Single color image to be detected, must be in RGB format.

Output

Detection result: A detection instance list. Each element in the list represents an identified object, containing its class, confidence score and standard rectangle box (non-rotated).

Parameter Description

Weight File

Parameter Description

Upload pre-trained YOLO model.

Parameter Adjustment

Supports both .onnx and .epicnn formats:

  • .onnx: Suitable for running on regular hardware devices.

  • .epicnn: Suitable for running on smart camera devices.

Enable GPU

Parameter Description

When using .onnx model, sets whether to use CPU or GPU computation. Note: This option is invalid for .epicnn models.

Parameter Adjustment

  • Enabled (recommended): For .onnx models, enabling in environments with NVIDIA graphics cards can greatly improve processing speed.

  • Disabled (default): Uses CPU for computation, slower speed but no special hardware requirements.

Confidence Threshold

Parameter Description

The model gives each detected target a confidence score, only targets with scores exceeding this threshold will be considered valid results and output.

Parameter Adjustment

  • Increase threshold: More strict filtering, can effectively reduce false positives but may increase false negatives.

  • Decrease threshold: More lenient filtering, can find more targets but may introduce more false positives.

Parameter Range

[0,1], default value: 0.8