Project Description
I want to make a model that can take a set of videos and an inputted goal, for example, a successful dunk, and delete the videos that did not capture the dunk and present the video that did capture the dunk. I thought about this idea because I often record my brother and sister trying to do tricks in their respective sports, and for me, I have multiple videos of me playing piano in my camera roll because I am too lazy to go back and listen to every video and see which one I successfully completed. I think that this technology would be useful for people who create digital content, but also for anyone, like parents, who like to record their children’s successes.
Success for this model would be to input a prompt and series of videos (with one video that fits the prompt) and the model correctly picks the right video. This project is definitely overly ambitious however it’s the first idea that came to mind because I know many people, parents especially, who have many repetitive videos of failed attempts on their phones, which takes up a lot of storage.
Failure Video…
Success Video!
Project Goals
- Choose one or two prompts to focus on
- Create a training dataset based on those prompts
- Train a model that is able to identify the video(s) matching the prompts