geo-11 TAB and SBS outputs produce some garbage pixels in certain resolutions
Posted: Wed Jul 20, 2022 3:34 pm
Hi, I was testing geo-11 on my 3DTV in its native 1080p res, as well as 2880x1620 (2.25x DSR). I was using TAB output and I noticed a weird line at the top of the image, only for one of the eyes. I was chatting with my buddy artox, who had the same issue and he found that it's gone when playing at 2160p (4x DSR).
So I went pixel peeping (Mirror's Edge: Catalyst and Snowrunner) and I think I found a possible root cause for the issue.
This was what I found when I took a look at a 1080p TAB screenshot (from the top row to the bottom-most):
540 lines of image for the one eye
4 lines of pixels which are almost identical copies (there's slight color shift) of the last row from the first image
536 lines of image for the other eye
Then I did the same for a 1620p TAB screenshot:
810 lines of image for the one eye
6 lines of pixels which are almost identical copies of the last row from the first image
804 lines of image for the other eye
I also checked a 2160p TAB screenshot - there are, as expected, 1080 lines for the first image followed by 1080 lines for the second one - no problematic rows.
Looking at these numbers, I came up with a theory that you guys are using a buffer/texture to store each view which has mod8 dimensions (540+4; 810+6). If my assumption was correct, this would mean that other resolutions which are mod8 would not exhibit this issue, so I tried 2560x1440 (720 lines per eye, which is mod8) and it worked without exhibiting the garbage lines.
I decided to check sbs with resolutions where the width of the image will not be mod8 but the stock DSR multipliers either result in odd-numbered resolutions or ones that are mod8. I tried DSR 1.20x - 2103x1183 (1501,5 x 1183 per eye?) which produced even weirder results at the border between both views.
I hope this will be easy to fix without reducing performance or other such potential drawbacks.
I'm attaching some raw screenshots and zoomed-in details for reference (taken from ME Catalyst), especially interesting for the messy sbs image.
_
So I went pixel peeping (Mirror's Edge: Catalyst and Snowrunner) and I think I found a possible root cause for the issue.
This was what I found when I took a look at a 1080p TAB screenshot (from the top row to the bottom-most):
540 lines of image for the one eye
4 lines of pixels which are almost identical copies (there's slight color shift) of the last row from the first image
536 lines of image for the other eye
Then I did the same for a 1620p TAB screenshot:
810 lines of image for the one eye
6 lines of pixels which are almost identical copies of the last row from the first image
804 lines of image for the other eye
I also checked a 2160p TAB screenshot - there are, as expected, 1080 lines for the first image followed by 1080 lines for the second one - no problematic rows.
Looking at these numbers, I came up with a theory that you guys are using a buffer/texture to store each view which has mod8 dimensions (540+4; 810+6). If my assumption was correct, this would mean that other resolutions which are mod8 would not exhibit this issue, so I tried 2560x1440 (720 lines per eye, which is mod8) and it worked without exhibiting the garbage lines.
I decided to check sbs with resolutions where the width of the image will not be mod8 but the stock DSR multipliers either result in odd-numbered resolutions or ones that are mod8. I tried DSR 1.20x - 2103x1183 (1501,5 x 1183 per eye?) which produced even weirder results at the border between both views.
I hope this will be easy to fix without reducing performance or other such potential drawbacks.
I'm attaching some raw screenshots and zoomed-in details for reference (taken from ME Catalyst), especially interesting for the messy sbs image.
_