Highlighted

Hyperlapse - Warp Stabilizer different results due to Analysis Server?

New Here ,
May 30, 2020

Copy link to clipboard

Copied

Hey,

 

I've been testing regular motion stabilization in combination with warp stabilizer hyperlapses for days now and experienced some weird behaviour. On one of my first tests I had great results with the default settings of warp stabilizer (Position & Rotation stabilization had been done in the childcomp). I kept on testing with the same footage and suddenly a new analysis showed different, shaky and warpy results. That was driving me crazy so I even started thinking If I remembered something wrong regarding the good results of my first testings.

In order to get again 100% I ran analysis multiple times and even started analyzing flipping/hopping tracking points one by one (even with detailed analysis checked) which can drive you nuts but I wanted to get 100%. I did get almost 100% by doing that but it had been really tough.

Today I tested with a different hyperlapse which wasn't wideangle, nothing distracting in the foreground. Position and rotation was fine so I ran warp stabilizer afterwards and nothing changed or it got even worse. Edges were shaky and so on. And then just like the opposite direction of my first clip I did multiple analysis and did again some position and rotation stabilization which looked the same as before but then suddenly warp stabilizer delivered great results out of the box. So I kept on changing position & rotation tracking in order to test if I could get bad results with warp stabilizer again but had no luck.

 

As I've seen AE is uploading stuff while analysing via „Adobe Analysis Server“ my question is if it possible to get different results with multiple analysis. It really drives me crazy if I do not really change a thing and suddenly it just works or stops working.

 

Here is the stable version of the second footage. I sadly didn't render the bad one.

 

Kind regards and thank in advance.

TOPICS
How to

Views

300

Likes

Translate

Translate

Report

Report
Community Guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more

Hyperlapse - Warp Stabilizer different results due to Analysis Server?

New Here ,
May 30, 2020

Copy link to clipboard

Copied

Hey,

 

I've been testing regular motion stabilization in combination with warp stabilizer hyperlapses for days now and experienced some weird behaviour. On one of my first tests I had great results with the default settings of warp stabilizer (Position & Rotation stabilization had been done in the childcomp). I kept on testing with the same footage and suddenly a new analysis showed different, shaky and warpy results. That was driving me crazy so I even started thinking If I remembered something wrong regarding the good results of my first testings.

In order to get again 100% I ran analysis multiple times and even started analyzing flipping/hopping tracking points one by one (even with detailed analysis checked) which can drive you nuts but I wanted to get 100%. I did get almost 100% by doing that but it had been really tough.

Today I tested with a different hyperlapse which wasn't wideangle, nothing distracting in the foreground. Position and rotation was fine so I ran warp stabilizer afterwards and nothing changed or it got even worse. Edges were shaky and so on. And then just like the opposite direction of my first clip I did multiple analysis and did again some position and rotation stabilization which looked the same as before but then suddenly warp stabilizer delivered great results out of the box. So I kept on changing position & rotation tracking in order to test if I could get bad results with warp stabilizer again but had no luck.

 

As I've seen AE is uploading stuff while analysing via „Adobe Analysis Server“ my question is if it possible to get different results with multiple analysis. It really drives me crazy if I do not really change a thing and suddenly it just works or stops working.

 

Here is the stable version of the second footage. I sadly didn't render the bad one.

 

Kind regards and thank in advance.

TOPICS
How to

Views

301

Likes

Translate

Translate

Report

Report
Community Guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
May 30, 2020 0
Guide ,
May 30, 2020

Copy link to clipboard

Copied

As far as I know, all calculations are done locally. I never noticed "Adobe Analysis Server", but server in generally doesn't always mean a remote computer. You can have a server on your local machine as well.

 

You can test your thesis if you disconnect the computer from internet and do the stabilisation. Do it again without internet and see if result is differnet. Do it again with internet and check results again. The double offline check is to make sure you get the same result everytime you ran the stabilisation without altering any parameter.

 

*Martin

Likes

Translate

Translate

Report

Report
Community Guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
Reply
Loading...
May 30, 2020 0
Adobe Community Professional ,
May 30, 2020

Copy link to clipboard

Copied

It is almost impossible for Warp Stabilizer to stabilize a hyper-lapse because most of them have way too much difference between frames. If the car coming down the street moves move than a couple of pixels Warp Stabilizer is not going to know what to do with the information. It is just too hard for the AI to figure out that the 300 X 600 block of pixels at 200, 500 is supposed to be the same object that is the 332 X 690 block of pixels that are now at 375, 550. The math doesn't look at the frames and say, OK, here's a bunch of pixels that are a car so let's look at the next frame and see if we can find the car.

 

The results you can get with Warp Stabilizer depend entirely on the shot, even with its regular video. Some shots just do not lend themselves to this kind of processing.

Likes

Translate

Translate

Report

Report
Community Guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
Reply
Loading...
May 30, 2020 0
Mylenium LATEST
Most Valuable Participant ,
May 31, 2020

Copy link to clipboard

Copied

Warp Stabilizer is not consistent and predictable. It's in the nature of the thing. Even tiny changes in the footage can result in the underlying motion field yielding a different result when transformed to abstract mathematical values, affecting the outcome. It's kind of a butterfly effect - tiny change, huge impact. Also note that the more you mess with this stuff manually the more you may actually work against it. "Over-tweaking" may introduce hard thresholds and sharp spikes in the underlying transforms where broad, smooth curves may be required. And of course this may exacerbate if you use the AI-backed functions - the AI may simply "over-learn" as each pass becomes part of its database and it may try all too hard to further improve results while looking for something new. It's like a human staring too long at something and beginning to see patterns when actually none are there...

 

Mylenium

Likes

Translate

Translate

Report

Report
Community Guidelines
Be kind and respectful, give credit to the original source of content, and search for duplicates before posting. Learn more
Reply
Loading...
May 31, 2020 0