I'm trying to build an engine for the following task:
I have n videos, from which I've taken 1 snapshot each. I am trying to train a classification algorithm on these n snapshots. Till now I have only tried to train a basic convolutional net but I think maybe their might be a problem with this approach since all snapshots are closely related, ie. there isn't a clear distinction between all categories.
Is there any other methodology available out there that could make this task simpler?
EDIT: My basic task is to build an app that would play the video associated to the image that is clicked by the user
EDIT: Okay so the problem is that I have a certain number of videos. I am capturing 1 frame per video. Say we give a certain user these, say n frames, 1 for each video. Now when the user captures a certain image, I want to play the video the image is a snapshot of.
My question is: Can this be considered as a classic classification problem? Or is it more of an image matching problem, like what we do in a Siamese network?