Comparing Scan-out high speed camera recordings

[dervu]

I have QD OLED 4K 240Hz screen and 1000fps camera and I would like to make comparison between two "setups" where theoretically there should be no difference in displayed image using following test: https://www.testufo.com/scanout

What approach would be best, and what should I look for?
Is it possible to sync frame numbers on left side to have same colors on same frame between two test runs (running test after restarting browser/PC)?

When I get to sync two videos, what should I look for to find any inconsistency in scanout? I guess timing would be most important, so I should be looking for frames where difference between two runs would be big, but how big is too much to be within limits of properly working scanout?

[dervu]

I performed a frame-skipping test and conducted two runs with slightly different setups, which should not affect the outcome.
It was necessary to restart the system between the two setups, but software and hardware conditions were the same.

I will confirm this behavior with additional runs to ensure it is consistent between the two setups, but I want to ask if QD-OLEDs are consistent enough that a square highlight pattern should not differ between two consecutive recordings for the same frame? (1000fps recording).

Videos were synced using first frame where first square was highlighted in same pattern.
Left side is first recording, right side is second recording.

• First setup comparison for same frame between two recordings.
  As you can notice, it is highlighted in different way. It goes like that for all frames, it is very rarely the same highlight pattern for each frame.
  
  
• Second setup comparison for same frame between two recordings.
  Here highlighting for squares is exactly the same between two recordings for each frame.
  
  

[dervu]

phpBB [video]

Is this normal variance between two on scanout test? Both synced on led light at click.

[Chief Blur Buster]

[Fixed your youtube to be an embed]

Yes.

Yeah, camera sensor scanout can be an error margin with some tests.

To avoid this; use landscape camera and back away far away from the screen, so that height of two screens is only ~10%-25% of camera frame height. This minimizes camera sensor scanout error margin.

(BTW, frameskipping, while you used it to correctly show camera sensor scanout behaviors -- is also not supposed to be used with high speed cameras, but the Scanout test (yes), is definitely designed to be used with high speed cameras. (For frameskipping you need to use long exposure (multiple refresh cycles) when using the Frameskipping test). Different TestUFO's do benefit from high speed camera (e.g. www.testufo.com/scanout ...) -- It is necessary to use the correct camera settings for the correct TestUFO test...

Now, your YouTube does show the scanout test, and you're conducting the test properly. But yes, 1000fps cameras usually have a 1/1000sec scanout. Use brute oversample (e.g. 4000fps Phantom Flex) or film from a distance (e.g. both displays use only 10%-25% of frame height). You can use post-editing to crop-zoom your result. Problem is, if you're using that thin ultrawidescreen slice, you're going to use only a few scanlines. It is quite correct that the bottom scanline of your 1000fps highspeed video frame is captured nigh nearly 1/1000sec after your top scanline.

Now, your high speed camera is an older one with a thin slice. Why not upgrade to a cheap Sony Experia XZ/XZ2 Premium or Galaxy, with that 960fps sensor? They produce vastly superior to older 1000fps consumer sensors, for a really cheap price, while being able to do HD (720p-1080p), and full frame height (making the 10%-25% height much less difficult). I've seen older-but-true Sony 960fps-sensored used smartphones for as little as $100-$200 off eBay, perfect for use with scanout test. That's only 10-20 cents per fps for a cheap HD high speed video camera that tends to work better for DIY scanout demos;

[Chief Blur Buster]

but I want to ask if QD-OLEDs are consistent enough that a square highlight pattern should not differ between two consecutive recordings for the same frame? (1000fps recording).

Error margin is always 1 frame. That's because the scanout-start of camera sensor is not time-synchronized to scanout-start of display panel. Necessarily that means session-to-session differences.

If you're concerned about session-to-session consistency on an atomic per-frame basis, you'll need all the following:
(A) Use only a tiny % of your visible frame height (e.g. 1/10th of your visible frame height)
(B) And/or a gigantically brute oversample margin (~10x-20x) for high speed camera framerate versus display refresh rate.