Help me find the right technique - Concatenative Video Synthesis?

Deep Learning
1

Hi all -

Halfway through the first course, I’m trying to find the right words to describe what I’m trying to do.

I’d like to take a video and replace a certain aspect of it with its nearest approximation in a separate corpus of video.

For example, I’d like to have a video of someone, identify their arm, and replace it with a gorillas arm, taken from many videos of gorillas etc.

I understand the first step would be an image classifier that identifies the subject to be replaced, but I’m not sure about the next steps. Is this concatenative synthesis, or something else? Has anyone seen this kind of thing implemented before?

1 Like
2

This looks like it would be a good segmentation problem.

1 Like
3

Hi Yorke hope your having a jolly day!

I am halfway through the course as well, although I have watched all the videos a few times making apps for each of the lessons means that some the lessons take me months not weeks. currently trying to deploy my 7 of my apps on Docker for Mac. :sweat_smile:

I plan to create an app like you describe as my final :trophy: project to celebrate completing part 1 of the fast.ai course Practical Deep Learning for Coders .

As ilovescience Help me find the right technique - Concatenative Video Synthesis? says it looks like a segmentation problem.

I also saw a video in one of Jeremy’s https://forums.fast.ai/u/jeremy lessons a video showing how a horse is turned into zebra using something called GANS.
https://medium.com/@jonathan_hui/gan-whats-generative-adversarial-networks-and-its-application-f39ed278ef09

So I will be looking at both of these techniques to help build my app. :sweat_smile:

Hope this helps.

mrfabulous1 :smiley::smiley: