R
R
Rustam Salavatov2014-11-03 01:23:33
Neural networks
Rustam Salavatov, 2014-11-03 01:23:33

Can a neural network be trained to edit video?

Good afternoon.
There is an idea to do video editing automatically.
Such a scheme of work is possible.
Let's say:
There are already frames from the video, with parameters in the database: face size, color, what objects (cats, cars, silhouettes, etc.)
There are patterns based on films.
I can also train the network myself.
It is necessary to make a film out of these shots. The network itself, by patterns and experience, must determine their order.
What will the process look like, and is it possible?
Which network or algorithm can be used.
It is possible to use self-learning systems like this
habrahabr.ru/company/itseez/blog/206116

Answer the question

In order to leave comments, you need to log in

3 answer(s)
S
Sergey, 2014-11-03
Protko @Fesor

I strongly doubt that a neural network is needed here...
www.disneyresearch.com/wp-content/uploads/Project_...

L
lightcaster, 2014-11-05
@lightcaster

I think it's possible.
But it won't be easy. You need a lot of data: original fragments and final video.
It's easier, I think, to start with automatic photo processing.

M
Michael, 2015-01-11
@mikkab

you can automatically mount, but not all. where a connected plot with a semantic load is needed - it is too difficult and not obvious in terms of costs and results. Auto-editing is now video synopsis & summarization, based on the analysis of motion / changes in the frame, they are no longer just a motion detector, but an analysis of its change.
There are several projects with video templates, but the result of their work is more like a slideshow, there is no meaningful editing there, although faces / shapes are defined and placed, which is no longer bad.
Automation for film editing, and even for distribution, is practically unrealistic, since not a single AI has yet learned to make or analyze the semantic content. for example, there are several startups for analyzing and writing scripts, but somehow there is not much movement, or there are simply too many graphomaniacs :)
Templates for some moments can really be useful, but directly compose a film with meaning from them, and if also with sound / speech, it will be extremely difficult.
Nevertheless, automation and analysis of film editing exists and works, but the goals are slightly different.

Didn't find what you were looking for?

Ask your question

Ask a Question

731 491 924 answers to any question