AI

Non-Maximum Suppression for Object Detection

Hard
Computer Vision

Task: Implement Non-Maximum Suppression (NMS)

Non-Maximum Suppression is a critical post-processing technique in object detection pipelines. When object detectors like YOLO, SSD, or Faster R-CNN generate predictions, they often produce multiple overlapping bounding boxes for the same object. NMS filters these redundant detections by keeping only the most confident box and suppressing boxes that significantly overlap with it.

Input:

  • boxes: An array-like of shape (N, 4) containing N bounding boxes in format [x1, y1, x2, y2] where (x1, y1) is the top-left corner and (x2, y2) is the bottom-right corner
  • scores: An array-like of shape (N,) containing confidence scores for each box
  • iou_threshold: A float between 0 and 1 specifying the Intersection over Union threshold for suppression

Output:

  • Return a list of indices of the kept boxes, ordered by descending confidence score
  • Return -1 for invalid inputs (mismatched dimensions, invalid shapes, threshold out of range)
  • Return empty list for empty input

Algorithm Overview:

  1. Select the box with the highest confidence score
  2. Remove all boxes that have high IoU overlap with the selected box
  3. Repeat until no boxes remain

IoU (Intersection over Union):

The IoU between two boxes measures their overlap and is computed as the area of intersection divided by the area of union.

Examples

Example 1:
Input: boxes = [[0, 0, 10, 10], [1, 1, 11, 11], [20, 20, 30, 30], [21, 21, 31, 31]] scores = [0.9, 0.8, 0.95, 0.7] iou_threshold = 0.5
Output: [2, 0]
Explanation: Box 2 (score 0.95) at [20,20,30,30] is selected first as it has the highest score. Box 3 (score 0.7) overlaps significantly with Box 2 (IoU = 81/119 = 0.68 > 0.5), so it is suppressed. Box 0 (score 0.9) at [0,0,10,10] is selected next. Box 1 (score 0.8) overlaps significantly with Box 0 (IoU = 81/119 = 0.68 > 0.5), so it is suppressed. Final result: indices [2, 0].

Starter Code

import numpy as np

def non_maximum_suppression(boxes, scores, iou_threshold):
    """
    Apply Non-Maximum Suppression (NMS) to bounding boxes.
    
    Args:
        boxes: Array-like of shape (N, 4) with boxes in format [x1, y1, x2, y2]
        scores: Array-like of shape (N,) with confidence scores
        iou_threshold: float, IoU threshold for suppression (0 to 1)
    
    Returns:
        List of indices of kept boxes, ordered by descending score
        Returns -1 for invalid inputs
    """
    # Your code here
    pass
Lines: 1Characters: 0
Ready
The AI Interview - Master AI/ML Interviews