Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/ilirosmanaj/detect_kermit/llms.txt

Use this file to discover all available pages before exploring further.

rotate_images.py is a data augmentation helper that effectively triples the training set by creating three rotated variants of each Kermit training image — rotated 90° clockwise (right), 90° counter-clockwise (left), and 180°. Running this script before training gives the model a much wider variety of image orientations to learn from without requiring any additional raw footage.

​
Source code

helpers/rotate_images.py
import glob
from typing import Optional

from utils import print_progress
from PIL import Image

PATH_TO_IMAGES = '../data/images/train/kermit/kermit-train-images'


def rotate_image(image_path: str, direction: Optional[str]):
    image = Image.open(image_path)

    if direction == 'right':
        rotated = image.rotate(90)
    elif direction == 'left':
        rotated = image.rotate(-90)
    else:
        rotated = image.rotate(180)

    image_name = image_path.split('/')[-1].replace('.jpg', '')
    rotated.save('{}/{}{}.jpg'.format(PATH_TO_IMAGES, image_name, direction))


def main():
    # get all image names
    images = glob.glob('{}/*.jpg'.format(PATH_TO_IMAGES))

    for i, image in enumerate(images):
        rotate_image(image, 'right')
        rotate_image(image, 'left')
        rotate_image(image)
        print_progress((i + 1) / len(images))

    print('\nRotated {} images'.format(len(images)))


if __name__ == '__main__':
    main()

​
How to use

Run the script from the helpers/ directory: 
cd helpers
python rotate_images.py
Progress is printed to stdout as a percentage that updates in place. A final summary line reports how many source images were processed once the script finishes.

​
Output naming

The rotated copies are saved alongside the originals inside PATH_TO_IMAGES. The direction string is appended directly to the base filename. For example, an original file named movie1frame100.jpg produces three new files:
FileRotation
movie1frame100right.jpg90° clockwise
movie1frame100left.jpg90° counter-clockwise
movie1frame100None.jpg180°
The None suffix on the 180° file comes from calling rotate_image(image) without a direction argument, which leaves the parameter as None — Python’s string formatting then renders it literally as None.

​
Configuration

The constant at the top of the script controls which directory is scanned and written to: 
PATH_TO_IMAGES = '../data/images/train/kermit/kermit-train-images'
Edit this value to point to a different image directory if your dataset is stored elsewhere. The path is relative to the helpers/ directory from which the script is run.
Data augmentation via rotation helps the model learn rotation-invariant features, improving accuracy on real-world video where Kermit may appear at various angles or camera orientations. Tripling the dataset this way is a low-cost alternative to collecting more raw footage and is especially valuable when the original frame count is limited.
print_progress is imported from helpers/utils.py. It writes an in-place percentage to stdout using \r carriage returns. See the API reference for utils.py for details.

Build docs developers (and LLMs) love

Get started for freeTalk to us