Yeah, I thought that might be the case - that was one of the big changes from 2.7 to 3.0 after all. Using DCPI-3 results in a bit of a washed out look for me, though.
2 Likes
Yes I thought I’d seen the problem in 2.7 and weirdly it was working for me locally in 3.0 - I’m pretty sure I hadn’t changed the output to DCI-P3. Hmm very strange anyway regardless I will fix this properly so dont worry.
1 Like
Quick update: I’ve found and squashed quite a few colour bugs. Essentially I’ve gone for colour manipulation that smoothly changes the colour bars in the test 240p. This has lead me to stripping out a few transformations that I had taken out of the grade shader such as phosphor gamuts. I may come back to those but in the meantime they’re gone as with complex/correct gamma curves on linearizing.
Everything is looking a lot better now without such extreme values that I was using before. I’ll hopefully submit this all tomorrow.
4 Likes
Hopefully I’ll see some sort of improvement while using my SDR 4K TV in 4:2:0 Limited colour format. There’s no reason why I should be able to get HSM Mega Bezel Reflection Shader and CRT-Guest-Advance to look so gorgeous while having to resort to RGB 4:4:4 exclusively when using the Sony Megatron Color Video Monitor. Isn’t it the same RGB Phosphor triads that both shaders are simulating at the end of the day?
No setting in CRT-Guest-Advance gives me the “green” screen 'effect" that I get when using the “wrong” TVL with the Sony Megatron Color Video Monitor at 4:2:0 Limited for example.
1 Like
Let’s see but I doubt it I’m afraid - the colours weren’t that far off. Only when you skipped the phosphor transforms were you seeing any big problems. The problems I’ve mostly fixed were much more subtle ones in the darks.
Let’s see what happens though and then I’ll explain what I think is happening if it does.
1 Like
@Cyber just tested it over Parsec which uses a less agressive 4:2:2 chroma subsampling and as expected it destroys the image just as before. So to explain why this will be effecting the Sony Megatron more than other shaders:
First what is 4:2:0 chroma sub-sampling doing? If you look at this image from RTings you can clearly see:
Its halving the images colour resolution or in other terms removing 75% of the colour information. Removing 75% of anything is going to have an impact but its even more so with high frequency patterns that the megatron uses. What are high frequency patterns? Basically any pattern that rapidly changes across an image or to put in other terms a pattern that goes from one extreme to another over a pixel or two - a 2 pixel alternating black and white line pattern being a classical high frequency pattern.
So imagine you got the black white alternating pattern and instead of using black and white you used red, green and blue instead to create a 3 pixel RGB alternating pattern. Then image you use 4:2:0 chroma subsampling on it or in other words throw away 75% of the colour information. In vertical direction you’re actually going to be alright (because there is no high frequency pattern in that direction) but in the horizontal direction your RGB pattern now becomes RRB or GGB etc but with luminance preserved.
This is what is happening to the Sony Megatron.
The other shaders will exhibit the same behaviours if they use 100% masks. I suspect in your case they’re not using 100% masks and so the mask is taking up the colour of the pixel either side or whatever the algorithm does and so lowering the frequency of the pattern. Some kind of blooming, blurring etc will also have a similar effect by transfering colour information across pixels and so lowering the pattern frequency. Lower frequency patterns get hit much less by chroma sub-sampling/dropping resolution.
The final thing I’ll say is if you take a look at the RTings page below and look at the three test images under the title “When does it matter?”. Notice how the blue on red text and vice versa are greatly impacted but the white on black much less so? Thats because the white/black pixels are all about luminance which is preserved where as red, blue is more about colour.
What I will say is that you really shouldn’t do any detailed sub pixel work using chroma subsampling its just not going to work well.
3 Likes
Thanks for the detailed explanation. I’m not sure if CRT-Guest-Advance is using less than 100% masks but I do have my mask settings at 100%.
Now to be fair, my comparison of both shaders is not quite apples to apples as I use completely different scaling settings when using CRT-Guest-Advance via HSM Mega Bezel Reflection Shader, however just as recently as a couple days ago, I did use the same scaling settings as recommended for Sony Megatron Color Video Monitor with HSM Mega Bezel Reflection Shader but I don’t know if I tried different mask sizes that might have been equivalent.
I find this phenomenon very interesting and one that might warrant some further investigation.
Perhaps, @guest.r, @Nesguy, @HyperspaceMadness or @hunterk might want to chime in as to why they think I haven’t seen this behaviour when using CRT-Guest-Advance? Also the colour generally seems spot on, without any adjustments at least in the latest versions which set the colourspace as sRGB by default and my Phosphor Gamut settings are generally set to 0 as well with a couple exceptions for games like Ninja Gaiden - NES, where I just find that NTSC-J looks best or rather any other setting looks wrong (even without my display being professionally calibrated and me providing look-up tables).
No hue adjustments, no white point adjustments - white point at 6500K.
I basically reset the colour settings to default in my last couple updates to my shader preset pack and the results seem rather compelling.
I think I have to setup crt-guest-advance using similar viewport scale settings as Sony Megatron Color Video Monitor and see if I can reproduce this behaviour using different mask sizes. I’m using Mask 6 which is the RGB mask by the way.
It’s really not magic how colour compression works - you will be experiencing it and the reason why it’s not apparent to you is because you are reducing the image frequency in some way: masks, blurs, blooms, etc etc what mask are you using for example?
As for all the colour stuff you mention you seem to be implying that you need to change things to get the Sony Megatron to display correctly - you do not, you are needing to do it because you are trying to make up for throwing away colour information.
I have all my settings at 0.0 - the only reason I changed them for SDR is to get a slightly brighter image for most people.
Hue as a value has nothing to do with any of this and it’s correct to change the white point to that of the CRT you are simulating - it is incorrect to have a white point of 6500K for a lot of Sony PVMs and Japanese made titles.
3 Likes
Not to derail the current discussion, which is valuable, but on the subject of color temp, doesn’t using a 9300K temp always result in clipped blues? The display is sRGB, windows is using an sRGB profile (sRGB uses a 6500K white point). Retroarch and all applications assume a 6500K white point.
Colors always seem poorly saturated and clipped when using 9300K because it pushes the blues out of gamut on a 6500K display. What am I missing?
Furthermore, wouldn’t using any NTSC shader cause clipping as well, since sRGB only covers about 72% of the NTSC gamut?
2 Likes
I wasn’t necessarily saying that. I was saying that with the neutral (default) colour settings that I currently use with HSM Mega Bezel Reflection Shader which uses CRT-Guest-Advance everything looks good despite the colour format being 4:2:0 Limited.
When I used 4:4:4 Sony Megatron Color Video Monitor also looks good with neutral (default) colour settings.
However, when I use Sony Megatron Color Video Monitor with default settings with 4:2:0 colour format it doesn’t seem to look correct. Additionally, if I try to use certain TVL and mask combinations I get green screen and stuff that doesn’t occur when using 4:4:4.
I’m not changing anything to compensate or implying that that is what needs to be done. You did mention some bugs in SDR mode and @Nesguy spoke about some issues as well. Perhaps it’s the same thing I’ve experienced. I probably need to update and give it a check again.
I understand the science behind the way colour compression works and why it was invented and thought to be feasible to begin with. I still appreciate the clarification though.
I do use a tiny bit of bloom but I don’t remember setting it to 0 and having any sort of colour issues apart from the image becoming a bit darker.
As I mentioned previously, I’ve been using HSM Mega Bezel Reflection Shader a lot longer than Sony Megatron Color Video Monitor so perhaps all that is needed is a little more time but I still find this interesting with respect to the behaviour of both shaders when using 4:2:0 colour format, which is the best my display can muster at 4K60Hz.
Will do some more investigating.
By the way, I don’t use any hue adjustments in any of my current presets since my recent overhaul.
1 Like
Possibly you’re right and colours get clipped but at the end of the day the CRT you’re simulating does have a different gamut to the screen you’re running the simulation on and as you point out your screens colour output maybe smaller (than the simulated CRT) in some places resulting in clipping.
However it’s still right to move the colour temp to that of the simulated screen as on an output display that can reproduce all those colours then that will be most accurate.
It’s kind of like SDR vs HDR - with an SDR display youre clipping luminance much more compared to HDR where you’re clipping less (or none) of the simulated CRTs luminance.
What I will say is that my hands on experience of side-by-side comparison of my 9300K PVM and the Sony Megatron on my Eve Spectrum is that setting the Sony Megatron colour temp to match results in much closer colours to the PVM.
3 Likes
Maybe we should frame this a different way: I’d expect the Sony Megatron to behave this way when using 4:2:2 or 4:2:0 chroma subsampling. As in Im pretty sure there’s not a bug - it makes sense that the colours are all junk with that kind of compression.
Maybe in a way I’m saying with the above statement that I think its kind of a bug that other shaders are working with chroma subsampling. The question then is a subjective one of what is correct behaviour and what is not - I’d say they both are in their own ways.
1 Like
does the PVM show clipping in color bars when looking at 240p Test Suite?
2 Likes
Not that I’ve noticed no - I would imagine that would be a well known thing too if it did happen right?
As I say 9300K is the standard white point for those PVMs and I’m sure some professional over all its usage over the years would have pointed out that colours are getting clipped even by small amounts that you can only really detect with a 24bit image and a colorimeter.
The snes 240p test suite is probably a pretty poor way to measure that kind of stuff anyway as how many shades of each colour are you getting on screen - 16? I don’t know but these things were designed with film in mind where there is many more colours on screen at the same time where you noticed the gradients being messed up.
3 Likes
Righty ho so I’ve just submitted pull request for the Sony Megatron v3.2 that adds back in all the colour gamut transforms that I temporarily took out in v3.1 to fix some subtle colour bugs in the darks. Hopefully this should fix the big colour issue that @Nesguy and @GPDP1 were seeing when using SDR sRGB output. It’ll hopefully fix all the issues in the darks that @Nesguy pointed out a while ago too.
These issues were largely to do with using the standard gamma correction curve for the rec.709 on gamma input which clamp dark values. This was then causing havok when the gamma_in value was tweaked. I’ve added a ‘gamma out’ value but this should be avoided from being used apart from when the output colour space is DCI-P3.
I may fix this problem in the future by using the more correct but more expensive mono-curves.
3 Likes
Hey maybe you could get in touch with this guy so you don’t have to keep relying on phone shots.
3 Likes
Quite possibly - thanks for the link! I presume he or she has taken a high resolution photo and then just zoomed in on it with Premier or something. Looks good though doesn’t it - they’re using a “Nikon D850 and a Micro-Nikkor 200mm F4D macro lens” I wonder how much that costs? Dread to think 
I’ll start putting penny’s in the jar.
2 Likes
You know this is the same link I shared earlier today in the Calling all CRT Owners: Photos Please Thread.
I felt that thread would have been more appropriate than this one.
2 Likes
Oh, sorry! I wasn’t following that one and just thought of this when I saw the video. Feel free to delete.
No worries man. You shared it better anyway with a direct link to the video. Lol