I have tested several anomaly detection repositories source code, and most of them didn't work. I have some advice for the upcoming code

Essential

• Test it on a single purpose container like Docker (or you can just use Google Colab if you want). A base image like Ubuntu 18 is a good start to track if there are dependencies that must be specified in a requirements.txt file.

• Prefer to use pathlib over os module or raw string for paths.

• Use absolute paths over relative paths to allow the scripts to run wherever the user is executing them.

• Use argparse to input paths (e.g. paths of directory). The following is a good example

import argparse
import pathlib


def parse_cli() -> argparse.ArgumentParser:
    current_directory = pathlib.Path(__file__).resolve().parent
    parser = argparse.ArgumentParser(description="Extract frames from directory containing videos.")
    parser.add_argument("--root", type=pathlib.Path, help="directory containing videos to be processed",
                        default=current_directory / "sample_video")
    parser.add_argument("--ext", type=str, help="extensions of the videos within the directory to be processed",
                        default="mp4")

    return parser

• Do not use something like for i in range(1, 101) to iterate video files. If a video doesn't have a name like 1.mp4, this won't work. Iterate them like pathlib objects and convert it to str when using OpenCV.
• Store data in JSON format over plain txt (unless it is just a list of strings like paths) or pickle files.
• If it will have links to download stuff like pre-trained weights or models, have different server options in case they get broken at some point. For example: Google Drive, Dropbox, etc.
• Add a Python gitignore to avoid committing unnecessary stuff.
• Use an IDE like Pycharm to inspect code, reformat it and catch some early bugs if possible. It's helpful to ensure that your code has some style.
• Measure the time a script takes to run

import time

def main(args: argparse.Namespace) -> int:
    start_time = time.perf_counter()

    # Some code
    
    end_time = time.perf_counter()
    res = time.strftime("%Hh:%Mm:%Ss", time.gmtime(end_time - start_time))
    print(f"Finished in {res}")
    return 0

if __name__ == "__main__":
    main(parse_cli().parse_args())

The reason of encapsulating the running code within the main function, you can find it in main — Top-level code environment

Optional

• Use decord like a cv2 style for random access.