Non-maximum Suppression

Non-Maximum Suppression (NMS) is a crucial technique in computer vision, particularly in object detection, used to filter out redundant bounding boxes that overlap significantly. It helps in selecting the most relevant bounding box for each detected object, thereby improving the accuracy of object detection systems.

Method

1. Bounding Box Generation: Object detection algorithms, like YOLO or SSD, produce numerous bounding boxes around detected objects, each associated with a confidence score indicating the likelihood of containing an object.
2. Sorting: The bounding boxes are sorted based on their confidence scores in descending order.
3. Selection Process:
   • The bounding box with the highest score is selected and added to the final output list.
   • The Intersection over Union (IoU) is calculated between this selected box and all other boxes.
   • Any box that has an IoU greater than a defined threshold (commonly set at $0.5$) is removed from consideration.
4. Iteration: This process repeats: the next highest scoring bounding box is selected, and the IoU calculations are performed again until no boxes remain.

IoU

Intersection over Union (IoU) is defined as the area of overlap between the predicted bounding box and the ground truth bounding box divided by the area of their union. Mathematically, it is expressed as:

$$\mathrm{IoU}=\frac{\text{Area of Overlap}}{\text{Area of Union}}$$

This metric ranges from $0$ to $1$, where $0$ indicates no overlap and $1$ indicates perfect overlap.

And the basic IoU loss can be expressed as:

$${\mathcal{L}}_{\text{IoU}}=1-\mathrm{IoU}$$