CODEWITHSUNDEEP
pythonalgorithmopencvobject-detection
LotaMaster
LotaMaster

Reputation: 15

OpenCV/Python - Finding Outliers by Area of Bounding Box

I have an object detection algorithm set up using TensorFlow, is there a way to remove the outliers in terms of the size of the boxes?

For example, I have 20 objects detected. Let's say 17 of them are around 50x50. But, there are a few bounding boxes that are 1x1 and one box that is 1000x1000. Obviously the 1x1 and 1000x1000 boxes are way too big and should be removed.

Upvotes: 1

Views: 857

Answers (1)

Fred Guth
Fred Guth

Reputation: 1667

One way you can do that is to use the z_score. The z_score will check how many std_devs does this number differs from the mean.

Example:

# coding: utf-8

import cv2
import numpy as np


bboxes = [(100,200), (120,210), (114, 195), (2,190), (104, 300), (111, 3), (110, 208), (114,205)]


def z_score(ys):
    mean_y = np.mean(ys)
    stdev_y = np.std(ys)
    z_scores = np.abs([(y - mean_y) / stdev_y for y in ys])
    return z_scores

thresh   = 1
outliers = [(t[0]>thresh or t[1]>thresh) for t in z_score(bboxes)]

This will print: [False, False, False, True, True, True, False, False]

Upvotes: 1

Related Questions