What does /vr/ think?
Me personally thinks hd-upscales with epsx/pcsxr look good in screenshots - for 3d games - but are atrocious ingame because of the jitter, despite gte accurancy. They are a no go for 3d games with prerendered bgs and 2d pixel art games.
Both widescreen hacks in pcsxr or mednafen suck, because they cause slow down and frame-drops and mess up 3d games with prerendered bgs.
Mednafen is very good and accurate. I like it. But it looks really low-res. It also does not support antialiasing.
I like to enhance the picture of mednafen for 3d games & 3d games with prerendered bgs with a bicubic filter. A) for smoothing the edges (antialiasing) and b) for smoothing textures.
For 2d pixel-art games I prefer a PVM, a good CRT or a good CRTshader. I think 3d games look so so on any kind of CRT/PVM device or simulation of such, because the degrading outweighs the enhancing.
2D games looks superior through a CRT because the textures/pixels are smoothed+bleed together to a certain amount through the CRT/analog input but appear really sharp again through the spacing of scanlines which looks really gorgeous in my opinion. 3D PS games on the other hand have prominent jagged edges, which are not meant to be there. So naturally you want to smooth the source to create a superior output, but not add scan lines, as they would degrade the picture again, making the jagged edges prominent again, instead of masking them.
I hooked up my PS to my CRT and my HDTV & found the output to be okay on my CRT, a bit too sharp - prominent jagged edges. The output on the HDTV was too blurry & smudgy for my taste.
I tried to mimic the effect of the PS1 on my CRT but without the scanlines & came to the conclusion that bicubic filtering is the best way to enhance native PS.
You see 3 pictures:
1 Raw
2 Bicubic MitchellNetravali (scientific best way to upscale an image psychovisually)
3 Custom Bicubic B-Spline filter
I personally prefer B-Spline- with the compromise of sharpness.
1. Mednafen with a front-end and accurate filters.
2. A PS2 with its enhancement features on.
3. A PSX with the best output you can mod it to support and a high-quality display (anything but LCD).
>>2890908
In that order, BTW. Hell, it's also prolly from least to most expensive too, but I can't be entirely sure for a given geographical pricing market.
Here is a comparison of the Bicubic Catmull-Rom filter and the Bicubic B-Spline by way of illustration.
Now imagine this filter in effect in for example games like Final Fantasy IX and its prerendered backgrounds.
This filter may be blurry, but it makes textures really smooth and the output interpolation really consistent instead of uneven, which I value the most.
>The best way to play Playstation games acurrately and as good looking as possible
PS1 > RGB > XRGB 3 > transcoder > High quality LCD.
Or maybe mess with emulation. Probably get something decent.
>>2890901
Why is integer ratio so hard for you idiots to understand? Look how shit the dithering is in your first image. Nearest neighbor is unfairly considered ugly because people keep doing it wrong.
Kinda getting tired of this whole what's the best way to play x meme. I just play with my old console and my old crt.
>>2890920
Those images are all integer scaled, I can assure you.
>>2890930
Not him and I don't know what the emulator is doing but your screenshot does look at least zoomed.
>>2890930
Then WTF is this? Your image, cropped and zoomed 10x. I assure you the PS1 did not output variable sized pixels. All those dots are supposed to be the same size. You fucked up the scaling.
>>2890915
My favorite general purpose upscaling filter is sigmoidized EWA lanczossharp. In imagemagick:
convert ff9.png -colorspace RGB +sigmoidal-contrast 5 -filter LanczosSharp -distort Resize \!512x256 -sigmoidal-contrast 5 -colorspace sRGB ff9_r.png
Note that the original image has some sharpening, which exaggerates the haloing no matter what filter you use. But look how much clearer the text is with this scaling, and the great balance between smoothness and sharpness.
You can do the same thing in real time with shaders in MPV, so I suppose somebody could make it work with emulators.
>>2890901
Perfection tier:
>original hardware + SD CRT TV
>accurate emulator input into SD CRT TV
High tier:
>accurate emulator + CRT monitor + shader
Those are the only way to really get high quality image accuracy.
4K emulator CRT shaders are okay but still not great. Anything under 4K is going to look bad.
>>2890943
>>2890956
Sorry, this is raw mednafen playstation retroarch 4x integer scaling.
I don't know, I guess this is an emulator or game specific bug, but it is integer scaling with overscan cropping turned off. Don't blame me.
320x240 x4 = 1280x960
>>2890901
>What does /vr/ think?
I think the Mednafen shilling needs to stop.
>CRT, shaders, filters
Literally autism: the post
Most of us just pop a PS1 disc into our real hardware and actually play the games instead of arguing over which method is subjectively "better."
>Amnesiac Kid
And this is why forced anonymous should be a thing on all boards.
>>2890983
CRUENTO PESTIS SHATRUEX PRAYANAVITA DOMUS-BHAAVA
>>2890967
Thank you. Really interesting.
Do you have the parameters for this filter? You could create one yourself.
CG Shaders are easily implemented. They usually follow a pattern like this:
Jinc2:
#pragma parameter JINC2_WINDOW_SINC "Window Sinc Param" 0.44 0.0 1.0 0.01
#pragma parameter JINC2_SINC "Sinc Param" 0.82 0.0 1.0 0.01
#pragma parameter JINC2_AR_STRENGTH "Anti-ringing Strength" 0.5 0.0 1.0 0.1
#ifdef PARAMETER_UNIFORM
uniform float JINC2_WINDOW_SINC;
uniform float JINC2_SINC;
uniform float JINC2_AR_STRENGTH;
#else
#define JINC2_WINDOW_SINC 0.44
#define JINC2_SINC 0.82
#define JINC2_AR_STRENGTH 0.5
#endif
Lanczos2:
#pragma parameter LANCZOS2_WINDOW_SINC "Window Sinc Param" 0.5 0.0 1.0 0.01
#pragma parameter LANCZOS2_SINC "Sinc Param" 1.0 0.0 1.0 0.01
#pragma parameter LANCZOS2_AR_STRENGTH "Anti-ringing Strength" 0.8 0.0 1.0 0.1
#ifdef PARAMETER_UNIFORM
uniform float LANCZOS2_WINDOW_SINC;
uniform float LANCZOS2_SINC;
uniform float LANCZOS2_AR_STRENGTH;
#else
#define LANCZOS2_WINDOW_SINC 0.5
#define LANCZOS2_SINC 1.0
#define LANCZOS2_AR_STRENGTH 0.8
#endif
Bicubic:
float ax = abs(x);
const float B = 1.0 / 3.0;
const float C = 1.0 / 3.0;
>>2890981
Fixed it for you, by reversing your defective integer ratio scaling, redoing it correctly, and then fixing the aspect ratio with horizontal axis only sigmoidized EWA lanczossharp. Flick between the two images and check out how much better the dithering looks now the pixels are all the same size.
>>2891007
Check out http://www.imagemagick.org/Usage/filter/nicolas/
Filter quality is somewhat subjective, so you might find something that's a better match for your eyes+display.
>>2891016
Can you tell me why the mednafen or retroarch is scaling it incorrectly?
>>2890983
Hey, I like you. Just hope people on this board remember how to use sage correctly
>>2891028
Sorry, I don't use mednafen so I don't know.
>>2891025
You are interesting. I like you. I usually linger on this site: http://www.imagemagick.org/Usage/filter/
But thank you, I' m gonna check Robidoux out as well.
>>2890974
Forgot to rename her to Aerittthhth
>>2891038
Aerisu.
>>2891040
Earisu?
>>2890983
Why do you hate Mednafen so much? It has the most accurate PCE and PSX emulation that Ryphecha wrote from scratch.
>>2890920
Integer ratio is almost impossible to guarantee on PSX, because games can use almost any resolution between 256x240p to 640x480i, and can change their resolution on the fly.
>>2891035
I found the bug. Ironically Retroarch applies the wrong core provided aspect ratio when overscan cropping is turned off. The only way to display the correct aspect ratio is to turn overscan cropping on. Thank you for bringing this to my attention. That it very unfornunate indeed. But now it is correct.
>>2890981
>>2891016
>>2891028
>>2891035
The game is obviously using 512x224/240/256 depending on the region and coding. It's not 320x240 natively so it's not going to scale up well unless you do something like 1024x896/960/1024.
Of course the intended aspect ratio would always be 4:3, but the most correct way to do an integer scaling of something odd like 512x224 while keeping a close ratio would be doing a vertical doubling first.
>>2891058
Btw this also goes for SNES games.
>>2891028
Mednafen simply outputs 4:3 in most cases. If you only did 1:1 pixel aspect ratio, then you would have games suddenly changing the display aspect ratio when they changed to a different resolution.
It's pretty much impossible to display PSX 4:3 and still always have integer scaling.
>>2891049
You're going to regret asking that, methinks.
>>2890901
Just curious, would perspective correct texture rasterization always improve a PSX frame? I can't imagine many games would rely on the incorrect screen coordinate rasterization.
>>2891068
Nah it's still wrong. It's only doing a somewhat better job at interpolating.
>>2891078
You mean the dither should be perfect squares? Gonna check to make that work.
I still use my PSone on my CRT with an S-video connection.
>>2891064
You can use a CPU filter to scale 3x integer nearest or so (I don't think one's included in the official distribution, but it's easy to write), and then bicubic or something in a shader. This is a reasonably good compromise between sharpness, speed, and artifacting.
>>2891081
Perfect rectangles actually, because the pixel aspect isn't 1:1.
>>2891081
Scratch what I said in >>2891078, and no I don't mean pixels should be perfect squares if you don't want the pic to get large (but that would be the closest to integer anything). I just mean every pixel has to be the same width.
In this case, all you have to do is get the 512x224 framebuffer and double it to 512x448 using nearest neighbour. And there it is.
>>2891078
You can't expect perfect square pixels on PSX, because games can use wildly non 4:3 resolutions internally.
Best recourse is to use sharp-bilinear or pixellate shaders if you want to keep pixels sharp without unevenness.
This game uses 512x240 during gameplay, but displays 320x240 during loading cutscenes and 640x480i in the main menu.
>>2890901
Thoughts:
1. If you're worried about accuracy, use real hardware
2. Harry Potter? You're gay
>>2891092
>You can use a CPU filter to scale 3x integer nearest or so (I don't think one's included in the official distribution, but it's easy to write), and then bicubic or something in a shader.
That's more or less what sharp-bilinear shader does, it prescales 2x to 5x nearest neighbor, then scales using bilinear to the screen resolution.
If you want to do it with something other than bilinear, you can make a shader preset that uses the stock shader to scale 2x to 5x Nearest, then put the interpolation shader in the last pass to scale to screen dimensions.
>>2891096
Why would they render at such a wide resolution? What's the point?
>>2891096
Does it switch automatically resolution or does mednafen/retroarch all output them to 320x240?
>>2891096
>sharp-bilinear
Haven't tried this, sounds good
>pixellate shaders
Tried this, it sucks because the sharpness varies depending on how the emulated pixels line up with the screen pixels.
>>2891119
The internal resolution changes with the game, but they always output as 4:3 in multiples of 320x240. I don't think there's an option to always force 1:1 PAR in Mednafen.
>>2891114
Probably for anti-aliasing purposes.
>>2891148
CRTs will display it as full screen 4:3, exceptthat CRT's don't scale anything and they simply display what the console sends to it, so they never have uneven pixels.
>>2891148
It always displayed 4:3.
Another complication - the PS1 could move the image on CRTs by adjusting the video signal timing. Some games had this in the options menu. And a few games (Chrono Cross was one IIRC) used this feature to do screen shaking effects. This literally moves the image by fractional pixels. Can any emulators handle this?
>>2891074
>I can't imagine many games would rely on the incorrect screen coordinate rasterization.
I can't imagine any developer building 3D assets around affine transformation. For static sprites and all that there are specific 2D functions which are rasterized without any vertex information.
people that make threads like this end up playing the game for 15 minutes
>>2891160
>and would get the same dither pattern as here
Yeah but that wouldn't stop every pixel from always keeping the same width. Not the case with your image there.
>>2891168
Exactly, CRTs are analog so they can have whatever shape pixels you like.
>>2891168
How so? Please explain it to me so I can understand- I don't get it. Stretching is stretching I though, if done by CRT or integer scaling in the emulator.
And what do you mean by same width? All the pixels in the image have the same width. The pixels are not square but they have the same width. If the internal resolution is not 320x240 the pixels can't be square, nor can the displayed signal on a CRT line display them that way. If you were to stretch the pixels to rectangles the image would become very distorted as seen in the attached image.
But maybe this is all wrong, as we don't how mednafen works. Maybe the resolution is displayed correctly just the dither pattern is displayed or scaled incorrectly which maybe a emulator or core specific bug.
>>2891183
Okay but isn't it as follows:
Ps generates let's say 512x240 image.
>CRT stretches it 4:3 aspect ratio in analog signal with lines. The pattern of a square displayed will be stretched accordingly.
>Emulator stretches image to 640x480 and the interg scales it or integer scales it and then stretches the image.
You end up with a non square pixel.
Maybe I'm just misunderstanding what you are trying to convery, because I totally get it wrong.
>>2891168
>Not the case with your image there.
The pixels are the same width there, you're looking at the wrong image.
>>2891197
That would explain the misunderstanding.
I want to play Suidoken II without throwing down 100+ ameribucks on it, so I plan on emulating it I guess
What emulator should I use? Any psx emulators with filters that could give an effect similar to my XRGB mini?
>>2891186
See this? I vertically doubled >>2891096 so that I could get the most correct integer scaling while keeping a close aspect ratio of what it's supposed to look like on a TV. To be even more accurate I would have needed to add artificial scan lines instead of doubling them but you get the point.
>>2891197
No they're not, do you want me to point out they aren't?
>>2891214
It's 5 dollars on PSN you poor cunt.
>>2891228
But a TV is stretching the image to 4:3 (1,333:1). The image you posted is 1,07:1. How can that be correct? It sounds theoretically correct, but in practice it's not accurate I guess. It would only be correct if the TV letterboxed non 320x240/640x480 content, but >>2891139 >>2891156 and >>2891158 said that it always is displayed as 4:3.
>>2891232
Oh shit, really? I haven't fired up my PS3 in a long ass time. Thanks for the info
>>2891242
I'm sorry, I don't wanna sound rude, but have you actually read the replies? It was said now at least three times that the improper scaling was caused by a emulator bug which was fixed with a setting here >>2891058
Please compare again
original version with wrong aspect ratio: >>2890981
your corrected version of the wrong aspect ratio: >>2891016
and the correct version with 4:3 aspect ratio: >>2891058
Please do review again everything you said in reply to this >>2891058 image and reconsider.
I think this was just a misunderstanding and you didn't see that the changed setting actually fixed the incorrect scaling and I thank you very much and I'm grateful that you brought it to my attention. That is all I have to say about this topic.
>>2891250
>by a emulator bug which was fixed with a setting here
And I hereby tell you that no it was not. Why? Because 512x240 just can't flatout be scaled with an integer factor if the resulting image is 1280 pixels wide, because 512*2=1024 and 512*3=1536, not 1280.
I have zoomed part of the image you're talking about and upscaled 2x. Excuse the poor drawings I made to illustrate my point.
>>2891272
That is unfortunate because mednafens internal resolution seems to be 320x240. There is no way to correct the dither pattern without destroying the intended aspect ratio of 4:3.
I could try it with a custom aspect ratio but it would basically stretch the 320x240 pixels to 512x240 or the multiples of it, but then the aspect ratio would be wrong.
>>2891313
AAAh okay! Gotcha!
You would need to use an output resolution of something like 2720x1920 at mininum to maintain integer scale at all possible PSX resolutions at an aspect ratio that's somewhat close to 4:3 (in this case, 17:12 or 1.41666). Higher integer scales of that could get you closer to 4:3 but you'd need like 8K or higher screens to display them.
Am I wrong for thinking S-Video for the PS1 looks wrong? Watched a video earlier today.
https://www.youtube.com/watch?v=oz-1zh4DjG4
Maybe it'll look sharper on my TV, I am not sure, but that "step up" just looks wrong to me.
>>2890901
1. Plug a PlayStation into a TV.
2. Play.
3. Enjoy. (Optional)
It's that easy!
PS1 games look great rendered at the original resolution but displayed at whatever your desktop is.
some of them look particularly good in widescreen, like Silent Hill and Metal Gear Solid.
Then again, most people can't stand the 'blocky' models because they're faggots. Anyone got those SH .GIFs?
>>2891332
This is a detail of retroarch/mednafen in 8K 7680x3840 letterboxed 4:3. Is the problem still prevalent? I can't tell. If so, it's not a resolution but an emulator/core/mednafen specific problem.
>>2891114
it's not as much a wide resolution, as it is a short resolution. Fewer pixels means less work for CPU and GPU. Especially for text, but also general imagery, we rely more on horizontal information than on vertical information. So instead of making the entire output proportionally smaller, it's smarter to use a distorted viewport with tall pixels.
>>2891148
CRTs distort the input image in an analog fashion. There's no integer upscaling and hence no rounding errors leading to uneven pixels. CRTs do not even have the concept of a pixel
>>2891160
CRTs don't use integer scaling, and have no concept of pixels. That's why uneven dither patterns can not occur. You can even change the size of the output image on your CRT (TV or monitor) with the various analog knobs, and it will only distort the image, but not introduce any scaling artifacts.
>>2891162
I don't know much about PSX architecture, but usually 2d stuff is just 3d with one of the coordinates zeroed out. If that is the case on the PSX, perspective correct texture mapping would be identical to screen coordinate texture mapping, as there is no perceptive to correct for. If there is special 2d hardware, the point is moot because you don't need perspective-anything in 2d.
>>2891313
There's a bit of a workaround I'm using on DOSBox: You upscale the internal resolution to 3x or so using nearest neighbor, then render that on screen using bilinear filtering. The result is a very slightly blurred nearest neighbor output. The advantage of the mild blurring is that misalignments with the output resolution are eaten up by antialiased pixel edges, instead of uneven pixel sizes
>>2891337
That is input into a capture card, not a CRT TV. It looks better in person.
>>2891062
Stretch scaling will actually reduce brightness on a real CRT.
Try for example turning between 4:3 and 16:9 on a 4:3 CRT. The latter will be brighter because same amount of light gets "drawn" on a smaller surface.
A real CRT filter should probably try and account for this during aspect correction.
>>2890908
> A PS2 with its enhancement features on.
What are those?
>>2891102
> 2. Harry Potter? You're gay
But it's fun game.
>>2891110
sharp bilinear doesn't handle interlacing
>>2890925
>still using a CRT when it's possible to make old games look amazing on an HD TV