Python Script for Face/Object Detection & Video Splitting

I have an existing object detection model which I used to process photos and videos from a specific sports event. I am looking to make some improvements with some added scripts that I can run to further process the data. Each one of these three should be able to be run independently. Specifically: 1. given an input folder of images, we will rename them by adding attempt number. This should also work on subfolders in the input folder. This will need to use clues such as who is in the image, the timestamps of the images as well as the positions of objects in the photo to make guesses about when a new attempt starts. the attempts should be added like 1att, 2att, etc to make it easy to know what the number means in the filename. 2. creating a local database of athlete ID. when a new athlete is detected, an id is assigned to it and stored it in database. if an athlete is recognized as previous one, we will get its id from database and change the filename accordingly. this should work on both photos and videos and it should also work on subfolders. it should detect any file that has been previously tagged and ignore it. the info added should be something like 2154id or some other recognizable text like id to make it easy to know what has been tagged. 3. for continuously recorded sessions, we will trim the videos according to the count of attempts present in the video. this will not crop the videos, just trim. it will use information from the ML model to find attempts. the trimmed video will be in the original format/codec/quality and it will use logic to process these files quickly, rather than evaluating every frame in a massive file. an example would be looking for lifts every second, rather than every frame, then narrowing in on the start of the lift more precisely when a lift has been found. All of these scripts should be resilient in the sense that they don't quit on themselves when something a bit weird is found. For instance, in part 1, it shouldn't quit because there's a series of photos not of barbells or athletes. It should just gracefully skip over it and keep going. Or for #2, same thing, if no one is found, it shouldn't just fail, it should probably just use the same person ID as the prior image if it finds that person in following images (for instance, let's say the model cannot see the face for an image or two, it should have logic in it to understand and adapt). And for number 3, it should not fail because a barbell is not seen or a person is not seen or similar, it should just keep looking. I want them to be more bug proof than the last script was.