INTERACT FORUM

Please login or register.

Login with username, password and session length
Advanced search  
Pages: [1]   Go Down

Author Topic: MadVR upscaling algorithms - can anyone suggest a good scene for comparing?  (Read 34424 times)

Sandy B Ridge

  • Citizen of the Universe
  • *****
  • Posts: 884

I've been playing about a bit with MadVR and different upscaling settings. Have you got a favourite scene or clip that you use to test any differences?

I would say you need something with a bit of fine detail eg text, some straight and curved lines of different slope and radii and some blocks or lines of different colours.

Is anime or animated videos particularly good for testing because of the well defined lines and blocks of colours?

What do you guys think? I'm just curious!

SBR
Logged

6233638

  • Regular Member
  • Citizen of the Universe
  • *****
  • Posts: 5353

It's really difficult to just suggest one thing, or even a couple of things to look at. I've spent a lot of time evaluating the various scaling options in madVR as they have evolved over time.

If your system can handle it, I would recommend simply using Jinc 3 with the anti-ringing filter enabled, but not the linear light option, for both Chroma and Luma upscaling. Never use linear light for upscaling.
Catmull-Rom with both the anti-ringing and linear light options enabled seems to give the best quality downsampling - this is the only option where linear light scaling is acceptable.


If your system can't handle that, or you are primarily watching a lot of highly compressed, low resolution videos, then things get a bit more complicated.

After Jinc 3 AR, my preference for Luma upscaling is SoftCubic 80. (without linear light/ar) As the name suggests, this gives a soft result, but it avoids introducing any ringing artifacts, tends to avoid aliasing very well, and can de-emphasise mosquito noise in poorly compressed content. (older DVDs for example) In fact, once we get per-resolution scaling options, I am likely to switch back to this for SD upscaling rather than using Jinc 3 AR. I watch a mixture of HD and SD content though, and Jinc 3 AR looks better with the HD stuff.

If that is too soft for you, you may want to try Lanczos 3 AR which gives similar results to Jinc but with more aliasing, and less of a performance hit, or Mitchell-Netravali which is in-between Lanczos and SoftCubic for sharpness, and performs very well. Both have a tendency to show aliasing and ringing though.

With animated content, I would definitely be leaning towards the softer scaling algorithms.


If you need to reduce Chroma scaling settings, I would try Lanczos 3 AR, then Bicubic 75 AR if your system can't handle that. Chroma scaling is quite well optimized these days though, so most systems should be able to handle Jinc 3 AR.


If you are having performance issues with downscaling, the first thing to try is disabling linear light scaling with Catmull-Rom. (turn that off before the anti-ringing filter)
Logged

Sandy B Ridge

  • Citizen of the Universe
  • *****
  • Posts: 884

Thanks, those are really good tips. Maybe JRiver could add your post as a wiki topic.

I've followed your excellent discussions on doom9 and this is a great summary.

I was just curious whether anyone has a few favourite clips/scenes to evaluate the different algorithms.

Cheers,

SBR
Logged

jmone

  • Administrator
  • Citizen of the Universe
  • *****
  • Posts: 14497
  • I won! I won!

Hey SBR,
If you have the re-released version of Gladiator on BD (note: the first BD version has big issues with so much processing that arrows would "disappear" mid flight) the opening war scene has lots of flying arrows etc which I find is a good demo.  That said I too also use Jinc3 +AR on the HTPC as my main focus is good BD playback.  On weaker GPU's I'd recommend Mitchell-Netravali as it is pretty good and is very easy on the GPU.  It gets down to preference at the end of the day, as you may not notice any great difference.
Thanks
Nathan
Logged
JRiver CEO Elect

jmone

  • Administrator
  • Citizen of the Universe
  • *****
  • Posts: 14497
  • I won! I won!

...also keep in mind that it really also depends on how big your screen is, your seating distance, and visual acuity.  If you have a 40" screen at 5m then just about anything will do.  It is much easier to see the impact of the algorithms on a big screen (say 100" at 3 feet)!
Logged
JRiver CEO Elect

InflatableMouse

  • MC Beta Team
  • Citizen of the Universe
  • *****
  • Posts: 3978

I picked a scene in which nothing was moving and something had closeup focus (sharp) with a busy, varying light and dark out of focus background. An example for this would be during a dialog in which the camera focuses on a face. I also compared different quality material like a good 720p, a good dvd and a bad dvd.

Jinc is typically the best but truthfully, I find it very hard to see it without making screenshots and comparing side by side with my nose close to the screen. And then when you do see a difference its not always a clear given which one is better. On my 50" plasma from 3 or 4 meters viewing distance I can't tell the difference anymore, so Lanczos it is for me. Also, personally I would choose a slight ringing or some artifacts over the softness of other methods, except for anime/cartoons.

I enable MadVR LAN Access features and change the settings from another PC, I find that easier than having MC drop out of exclusive mode to access it. I don't think settings are always applied on pause so to be sure I always restart and pause it on the same frame (which is frustrating ...). There's probably an easier way though but I couldn't be bothered to figure it out ::).
Logged

jmone

  • Administrator
  • Citizen of the Universe
  • *****
  • Posts: 14497
  • I won! I won!

Yup - at some point you have to kick back and enjoy the content.  For example I've seen my kids watch stuff on FTA with heavy snow and think it is fine!
Logged
JRiver CEO Elect

Sandy B Ridge

  • Citizen of the Universe
  • *****
  • Posts: 884

...also keep in mind that it really also depends on how big your screen is, your seating distance, and visual acuity.  If you have a 40" screen at 5m then just about anything will do.  It is much easier to see the impact of the algorithms on a big screen (say 100" at 3 feet)!
Ha! Yes I find I do have to move a little closer to the screen these days to clearly see the difference at pixel level.  However when I move back to the sofa there remains an appreciable difference particularly in sharpness between different settings. I have a 50" plasma at about 10-12 feet, so probably *should* be able to see a difference!
 
Hey SBR,
If you have the re-released version of Gladiator on BD (note: the first BD version has big issues with so much processing that arrows would "disappear" mid flight) the opening war scene has lots of flying arrows etc which I find is a good demo.  That said I too also use Jinc3 +AR on the HTPC as my main focus is good BD playback.  On weaker GPU's I'd recommend Mitchell-Netravali as it is pretty good and is very easy on the GPU.  It gets down to preference at the end of the day, as you may not notice any great difference.
Thanks. I'll try and get that one. I've been meaning to watch it.
I have to admit that I struggle to see the difference on bluray with the only reliable change I can detect between setting chroma to 'nearest neighbour' vs. anything else. Setting chroma to Jinc3 vs. Mitchell netravali for example I rarely can tell the switch!

SBR
Logged

Sandy B Ridge

  • Citizen of the Universe
  • *****
  • Posts: 884

Yup - at some point you have to kick back and enjoy the content.  For example I've seen my kids watch stuff on FTA with heavy snow and think it is fine!
+1

It is however innate in my nature to want to fiddle!

SBR
Logged

InflatableMouse

  • MC Beta Team
  • Citizen of the Universe
  • *****
  • Posts: 3978

My biggest annoyances during movie watching are banding and incorrect brightness/gamma/blacklevels.

I find the last one strange as I don't think this should differ from movie to movie, but it does. Black needs to be black and if it isn't, I get annoyed  >:(.
Logged

Sandy B Ridge

  • Citizen of the Universe
  • *****
  • Posts: 884
Re: MadVR upscaling algorithms - can anyone suggest a good scene for comparing?
« Reply #10 on: February 25, 2013, 05:32:42 am »

I picked a scene in which nothing was moving and something had closeup focus (sharp) with a busy, varying light and dark out of focus background. An example for this would be during a dialog in which the camera focuses on a face. I also compared different quality material like a good 720p, a good dvd and a bad dvd.
Good tip, thanks. Static scenes are probably really imporatnt for SD/DVD material otherwise you just end up upscaling macroblocks! I hadn't thought about faces before, so thanks for that. I had been trying industrial sort of scenes with fine text hoardings and metal lines.
Quote
Jinc is typically the best but truthfully, I find it very hard to see it without making screenshots and comparing side by side with my nose close to the screen. And then when you do see a difference its not always a clear given which one is better. On my 50" plasma from 3 or 4 meters viewing distance I can't tell the difference anymore, so Lanczos it is for me. Also, personally I would choose a slight ringing or some artifacts over the softness of other methods, except for anime/cartoons.
I find animated stuff sometimes a bit soft with softcubic, but haven't tried much with Jinc. Lanczos gets a bit blocky around the edges of lines at certain slopes/angles. I find Jinc a bit better at smoothing out these lines though, even with animated stuff.
Quote
I enable MadVR LAN Access features and change the settings from another PC, I find that easier than having MC drop out of exclusive mode to access it. I don't think settings are always applied on pause so to be sure I always restart and pause it on the same frame (which is frustrating ...). There's probably an easier way though but I couldn't be bothered to figure it out ::).
Really good tip. What I have done is program the colour keys on my remote to different upscaling algorithms. I don't have any use for them otherwise, so thought 'may as well'. The leftmost one is the softest and the rightmost one the sharpest. I can change the algorithm then on the fly.

Frame by frame advance would be a worthwhile feature enhancement for MC I guess.

SBR
Logged

JimH

  • Administrator
  • Citizen of the Universe
  • *****
  • Posts: 72536
  • Where did I put my teeth?
Re: MadVR upscaling algorithms - can anyone suggest a good scene for comparing?
« Reply #11 on: February 25, 2013, 06:57:13 am »

There was a black problem with Nvidia (?) recently.
Logged

eddyshere

  • Galactic Citizen
  • ****
  • Posts: 469
Re: MadVR upscaling algorithms - can anyone suggest a good scene for comparing?
« Reply #12 on: February 26, 2013, 01:43:16 pm »

Thanks, those are really good tips. Maybe JRiver could add your post as a wiki topic.

I've followed your excellent discussions on doom9 and this is a great summary.

I was just curious whether anyone has a few favourite clips/scenes to evaluate the different algorithms.

Cheers,

SBR

The opening scene in each episode of a tv show called frasier where a white line is drawn and represents Seattle...
Logged

Sandy B Ridge

  • Citizen of the Universe
  • *****
  • Posts: 884
Re: MadVR upscaling algorithms - can anyone suggest a good scene for comparing?
« Reply #13 on: February 26, 2013, 03:47:35 pm »

The opening scene in each episode of a tv show called frasier where a white line is drawn and represents Seattle...
Thanks for that. You're absolutely right. The multiple thin White lines at various slopes/angles would be really good.
I think it's only on a low bitrate SD channel at the moment on UK TV, so may not be that great. I'll keep an eye out though.

Thanks!

SBR
Logged
Pages: [1]   Go Up