APPLE FINAL CUT PRO: Apple Final Cut Pro X Creative Community Conversations FCP Legacy FCP Tutorials

Apple Silicon FCP now with Machine Learning.

COW Forums : Creative Community Conversations (was FCPX Debates)

<< PREVIOUS   •   VIEW ALL   •   PRINT   •   NEXT >>
Chris Ashworth
Apple Silicon FCP now with Machine Learning.
on Jul 3, 2020 at 5:46:10 pm

Here is what our next possible FCP might be... a version that integrates the neural engine in both the iPhone and Apple Silicon Computer (with its own Neural Engine) to do automatic tasks. Creating a tight integration with the iPhone Camera with the new FCP Editor with machine learning in the both devices Neural Engines an end to end solution to do these things automatically, XML Labelling, Object Recognition, Face Detection & Recognition, Depth Analysis with new (Lidar camera), Rotoscoping, Text to Speech captioning, Auto Colour Grading and Live Content Moderation. The average Joe will get an Editor that does all the heavy lifting when they shoot with an IOS device with the dual Neural Engine integration. So this will be a closed loop, android video will not talk to the new FCP version in the same manner as an iPhone.


Return to posts index

Ralph Hajik
Re: Apple Silicon FCP now with Machine Learning.
on Jul 3, 2020 at 10:15:13 pm

Hi Chris

Thanks for the information.

Safe & Happy Travels
Ralph Hajik
RJTravelMedia
http://www.RJTravelMedia.com


Return to posts index

Oliver Peters
Re: Apple Silicon FCP now with Machine Learning.
on Jul 3, 2020 at 11:24:30 pm

It's an interesting "what if" idea. But, don't hold your breath. As an FYI - FCPX has already had a basic level of this since day one. That is the ability to analyze and identify clips by wide shots, close-ups, etc.

Remember, "machine learning" does not actually involve any "learning." It's the software comparing the picture or sound against a supplied data set.

Pixelmator Pro uses machine learning to identify and automatically name layers. If you bring in an image of the Eiffel Tower onto a layer, it will generically name it "building" or "tower." Even if you correct the name to Eiffel Tower, it will not remember that for the future or learn from that in order to update the reference data set that it uses.

Do an image search on any internet search engine for a known person. The first batch of hits will be the correct person. However, then the images start to display a lot of similar-appearing individuals that are incorrect. That's because the facial recognition technology simply isn't that good.

Fortunately, it will still be a while before the robots take over ☺

- Oliver

Oliver Peters - oliverpeters.com


Return to posts index


Tim Wilson
Re: Apple Silicon FCP now with Machine Learning.
on Jul 5, 2020 at 10:50:31 pm

[Oliver Peters] "Remember, "machine learning" does not actually involve any "learning." It's the software comparing the picture or sound against a supplied data set."

Machines are stupid. LOL And the people who program, well, we'll see.



[Oliver Peters] "Do an image search on any internet search engine for a known person. The first batch of hits will be the correct person. "

This isn't my experience AT ALL. I think Google image search (the only one I use) is getting worse and worse, at least partly because it IS trying to learn, and data sets are becoming lower quality over time.

I have a couple of little hobbies like most of us do, and one of mine is posting about music, especially classic rock, over on tumblr. I don't always have time to indulge, but when I do, I want to post the details of a photo whenever I can -- I usually know who it is (even if I may not know who the OTHER person in the picture is), but I like to also credit the photographer, the date, the venue, all that kind of thing.

Look, I happen to know that this is Robert Plant, and I'm pretty sure that it's from the 1976 film, The Song Remains The Same.



For guys our age, Oliver, this is as mainstream as any music movie ever made, right? And Led Zeppelin is hardly obscure. But what does Google have to tell me?



Check the top of the results. Singer? Is that all you can tell me? Well, I guess "singer" is a good enough start...but the text results are for SINGER SEWING MACHINES!!!! Not even the right kind of singer!!!!

Looking up John Paul Jones images within the Led Zeppelin universe, the most frequent result is "girl". That's fine as a start. LOL But come on, this is maybe the second-most photographed band in the world after The Beatles, and the online fan community (Pinterest, Tumblr, Instagram, etc) posting pictures of them may be even larger, but with all this data Google can't figure this out?

This was definitely NOT always the case. When I first started doing this in 2013, I got useful results almost every time, or at least a clue where I could turn for more information. But I think that the combination of Pinterest and Instagram stripping metadata that's usable anywhere outside their platforms has made it harder for Google to follow along.

I also saw this image, and thought, hmmm, when did Robert Plant and Roger Daltrey recently hang out?



So I right-clicked, asked Google, and they told me that they had NO results, but they DID have a "best guess":



That's pathetic. I wound up going to TheWho.com, and searching for Robert Plant by hand, like a wild animal. LOL Turns out that this photo was taken at the 2011 dedication UCLA Daltrey/Townshend Teen & Young Adult Cancer Program, which Robert was there to support. Photo by Andrew Evans. C'mon Google, was that so hard?

I could be wrong about the mechanics of why Google image searches within classic rock are less useful over time, but I get that feeling everywhere I go. Good information is getting harder to find, and it feels like less and less useful search engines are closer to the problem end of the spectrum than the solution end.


[Oliver Peters] "Fortunately, it will still be a while before the robots take over ☺"

Me, when I think of AI, I'm thinking, "I don't care if it's artificial or not. Show me something ACTUALLY Intelligent." LOL


[Oliver Peters] "As an FYI - FCPX has already had a basic level of this since day one. That is the ability to analyze and identify clips by wide shots, close-ups, etc."

My snark notwithstanding, that's useful. I also like all of the language-related searches. Enter a text search term, and up pops not just anyplace I've entered that term into a metadata field, but also every clip where somebody speaks that word. I remember demoing the rollout of the product then known as Avid Interplay and doubleclicking a word in a transcript creating a subclip with that line in a bin.

A variation of this for news VO allowed me to delete words from a transcript, and the timeline would automatically rebuild itself to fit the slot that I'd allowed for the VO in the fixed-length segment, then either delete b-roll, or retime to stretch if there were no handles to lengthen the clips. That was in 2006, and it felt pretty dang intelligent to me.

So I'm not saying that there's NO place for AI in editing, but yeah, I definitely agree, Oliver, very easy for our enthusiasm for what COULD be possible to outpace the utility of what actually is possible.

But hey, we change what's possible when we wish for something better and start building it, right? :-)


Return to posts index

Oliver Peters
Re: Apple Silicon FCP now with Machine Learning.
on Jul 6, 2020 at 12:25:36 am

[Tim Wilson] "This isn't my experience AT ALL. I think Google image search (the only one I use) is getting worse and worse, at least partly because it IS trying to learn, and data sets are becoming lower quality over time."

It seems you are trying to find an identification starting from an image. I was talking about a regular old Google search entry and then seeing what associated images come up related to that entry.

- Oliver

Oliver Peters - oliverpeters.com


Return to posts index

Oliver Peters
Re: Apple Silicon FCP now with Machine Learning.
on Jul 6, 2020 at 12:37:02 am

[Tim Wilson] "I could be wrong about the mechanics of why Google image searches within classic rock are less useful over time, but I get that feeling everywhere I go. Good information is getting harder to find, and it feels like less and less useful search engines are closer to the problem end of the spectrum than the solution end."

There's plenty of research pointing to how bad facial/image recognition really is. One simple example is that AI is pretty good at identifying images as people - largely because of all the selfies on the net taken from roughly the same position. But it's much worse at identifying cats as cats. The theory being that images of cats come in all shapes, angles, and positions. Plus fur and color differences.

But in general - and pertinent to your dilemma - the internet is a terrible place for finding accurate historical information for anything that happened prior to the explosion of the internet. Most of that info only exists on the net through various archives and people who have posted the info as a passion on their own blogs.

- Oliver

Oliver Peters - oliverpeters.com


Return to posts index


Tom Sefton
Re: Apple Silicon FCP now with Machine Learning.
on Jul 6, 2020 at 10:58:06 pm

Does anybody remember laughter

Co-owner at Pollen Studio
http://www.pollenstudio.co.uk


Return to posts index


<< PREVIOUS   •   VIEW ALL   •   PRINT   •   NEXT >>
© 2020 CreativeCOW.net All Rights Reserved
[TOP]