Get object from bounding box [Object Detection]

Question

I have a .txt file that contains path/to/image.jpg,xmin,ymin,xmax,ymax for each row and a img folder that contains the jpg images. What is the best way using python to extract the "objects" inside the coordinates of each file and look if the bounding boxes are set correctly?

Thanks!

Answer 1

You can use the opencv-python library:

import cv2

# Reading the file
with open("filename.txt") as iostream:
    content = iostream.read()

# Visualizing the data
color = (0, 0, 255) # RED
for line in content.split("\n"):
    image_path, xmin, ymin, xmax, ymax = line.split(",")
    image = cv2.imread(image_path)
    pt1 = (int(xmin), int(ymin))
    pt2 = (int(xmax), int(ymax))
    cv2.rectangle(image, pt1, pt2, color)
    cv2.imshow("Visualization bounding box", image)
    cv2.waitKey()

This code will show each image, with a red rectangle to display the bounding box. If you press any key, it will switch to the next one.

If you want to save the cropped images, you can use something like this:

    outfile = image_path[:-4] + "_bbox.jpg"
    outimage = image[int(ymin):int(ymax), int(xmin):int(xmax)]
    cv2.imwrite(outfile, outimage)

Answer 2

Your text file is a CSV (Comma Separated Values) and can be loaded as such.

You could simply use PIL to crop each image in your folder to the bounding box:

import csv
from PIL import Image

f = open('test.csv')
csv_f = csv.reader(f)

#iterate through rows of your CSV
for row in csv_f:

  #open image using PIL
  im = Image.open(row[0])
  #crop box format: xmin, ymin, xmax, ymax
  crop_box = (row[1], row[2], row[3], row[4])
  #convert values to int
  crop_box = map(int, crop_box)
  #crop using PIL
  im = im.crop((crop_box))
  #save the image to predetermined savepath
  im.save(save_name)