🟠 Detection Result Split

Function Description

This operator is used to split image detection results, extracting each target’s mask image, center point coordinates, and corresponding detection box information.

Usage Scenarios

Meeting post-processing requirements based on detection or segmentation results.

Input Output

Input

Image: Supports input of a single RGB image.

Detection result: Detection output results, containing each target’s contour polygon, class, score and other information.

Output

Detection result mask image list: Binary mask image list corresponding to each target, where white areas represent target regions.

Center point list: Center point pixel coordinates of each target, N×2 array.

Detection box list: Detection box four corner coordinates of each target. Can be minimum bounding rectangle or extremum rectangle, format is N×4×2 array.

Parameter Description

Detection Box Type

Parameter Description

Used to select the method for generating target detection boxes, affecting the final output detection box type and accuracy.

Parameter Adjustment

  • Minimum bounding box: The minimum bounding rotated rectangle, which calculates a minimum rotated bounding rectangle for each detected object based on its precise contour points. This method can obtain more accurate detection boxes that closely fit the object contour, especially suitable for scenarios where objects have irregular shapes or subsequent tasks require high accuracy in object angle and size.

  • Maximum minimum point: This method simply takes the maximum and minimum values in X-axis and Y-axis directions based on the object’s contour points, directly constructing a horizontal rectangle box aligned with coordinate axes (i.e., top-left corner point and bottom-right corner point). It has fast calculation speed but lower detection box accuracy, and the box will contain more blank areas, especially when objects have large rotation angles or are relatively tilted, it may significantly exceed the actual object boundaries.