madVR - high quality video renderer (GPU assisted)

[madshi]

From my observation the actual difference in smoothness that different graphics exhibit w/ the current iteration of madVR stem only from (limited) memory bandwidth.
The current version of madVR does barely any shader math. That will change. I don't know, though, how high the math load will be in the end. It's possible that madVR will stay texture read limited...

[tetsuo55]

Yes it will. Those specs bring nothing new. The 30 bit they talk to are for the three components. madVR processing is 16bit per component, or 48bit for the three components. And there is also all other stuff that is/would be possible using the 3dluts.

I was being partly sarcastic. We know from testing that almost none of those features work as advertised.

This makes madVR extremely important.
If it was able to support DXVA it would be my default renderer.

madVR does support CUDA right? so getting an nvidia card would mean hardware acceleration?

[leeperry]

madVR does support CUDA right? so getting an nvidia card would mean hardware acceleration?
yup!**

[honai]

madVR does not interfere with CoreAVC w/ CUDA, so yes, with an nVidia card you'd get GPU acceleration of AVC streams.

With the current economic situation I don't see AMD/ATI developing something similar to CUDA in the near future, with developing being defined as actually shipping ...

[tetsuo55]

yup!**

But i forgot that only works for h264 so no luck with VC1 and MPEG2

[honai]

Typical VC-1 and MPEG-2 streams from Blu-ray require much less processing power than AVC. A 5-year old system can decode most MPEG-2 streams just fine.

[tetsuo55]

Typical VC-1 and MPEG-2 streams from Blu-ray require much less processing power than AVC. A 5-year old system can decode most MPEG-2 streams just fine.

I guess my system is older then (Athlon XP 2600+ @ 2,1ghz)

(I'm going for a completely CPU based solution anyway ,a 3ghz+ C2D, but it still need a videocard so i want to get the best one HTPC wise)

EDIT:

Wikipedia say's it's release date is September 21, 2003

[nlnl]

I was being partly sarcastic. We know from testing that almost none of those features work as advertised.

This makes madVR extremely important.
If it was able to support DXVA it would be my default renderer.

madVR does support CUDA right? so getting an nvidia card would mean hardware acceleration?

Yes, but Cuda-based decoding does load the GPU!
Madvr avrg GPU rendering time is up about 30% :(!
Have a look at my tests in this post:
http://forum.doom9.org/showthread.php?p=1278340#post1278340

[cyberbeing]

Yes, but Cuda-based decoding does load the GPU!
Madvr avrg GPU rendering time is up about 30% :(!
Have a look at my tests in this post:
http://forum.doom9.org/showthread.php?p=1278340#post1278340

Well if NVIDIA's GT300 really ends up being a MIMD part (instead of SIMD) (http://www.brightsideofnews.com/news/2009/4/22/nvidias-gt300-specifications-revealed---its-a-cgpu!.aspx), supposed 3TFLOPs nonwithstanding, this problem will likely be solvable with the release of that hardware in Q4 this year. In this way the CUDA code and the madVR code could be run in parallel, at the same time, without having to wait on the processing of the other like in the traditional SIMD architecture. Of course, this is likely another beast of a card, and something that will never be aimed towards the HTPC market. It still will be interesting if NVIDIA really does release a MIMD part before the Intel Larrabee which is also MIMD part.

[nijiko]

I just waiting for 0.9 that fixed the RMVB playback problem.
Is it probably?

[tetsuo55]

I just waiting for 0.9 that fixed the RMVB playback problem.
Is it probably?

Hi nijiko,

I do not have this problem with ffdshow and realalternative with MPC-HC.

I watched over 10 hours of rmvb without any problems

[leeperry]

BTW, good displays already do 10bit+.
I forgot this point. the major change between the Sanyo PLV-Z2 and Z3 was that the Z2 had a 8bit processing, and the Z3 10bit.

I've owned both projectors at the same time, and was able to compare them almost simultaneously...well, it was exactly like the Sanyo advertisment :

http://www.son-video.com/images/static/Rayons/HomeCinema/VideoProj/Sanyo/PLVZ3_PixelWorks_FX_m.jpg

the picture was grainy to death on the Z2, totally unbearable when you've seen the Z3(and when you've seen the Z4, you can't stand the Z3 screen-door effect anymore).

and my current DLP has "a combined full 10-bit processing I/P converter and scaler, and a 12-bit floating point digital gamma controller. The result is approximately four times the number of gradations of a conventional 8-bit model, so even skin tones are faithfully reproduced."

http://mitsubishi-electric.com.au/354.htm

but it doesn't cut down to 8bit again, you actually get all this resolution "on screen".

latest Sanyo's have full 12bit processing, mostly because they have a pretty crazy CMS....but I've read several times than anything above 10bit was overkill...you'll tell me that it's not, I guess.

PS: and about the stimulus thingie, Colorfacts won't let you measure the display saturations(Color.HCFR will, but we have no way to fix it from the PC anyway?) so if your saturations look like this, this is m00t IMHO :

http://www.image-load.eu/out.php/t159894_satu.png (http://www.image-load.eu/out.php/i159894_satu.png)

w/o a perfect display and a $2000 probe(freshly recalibrated please) in a pitch black room w/ black painted walls, you won't get any benefit from all this...but anyway it's great that you implement it!

[nijiko]

Hi nijiko,

I do not have this problem with ffdshow and realalternative with MPC-HC.

I watched over 10 hours of rmvb without any problems

The problem I can recur. So I confirm that's a bug.

The video info.


滤镜 : MPC - RealVideo Decoder - CLSID : {238D0F23-5DC9-45A6-9BE2-666160C324DD}

- 连接到:

CLSID: {765035B3-5944-4A94-806B-20EE3415F26F}
滤镜: ...\千年女优.rmvb
Pin: 输出 00 (Video Stream)

- 连接的媒体类型:

Video: RealVideo 4 1280x688 23.98fps 560Kbps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Video {73646976-0000-0010-8000-00AA00389B71}
subtype: Unknown GUID Name {30345652-0000-0010-8000-00AA00389B71}
formattype: FORMAT_VideoInfo {05589F80-C356-11CE-BF01-00AA0055595A}
bFixedSizeSamples: 1
bTemporalCompression: 0
lSampleSize: 21472
cbFormat: 122

VIDEOINFOHEADER:
rcSource: (0,0)-(0,0)
rcTarget: (0,0)-(0,0)
dwBitRate: 560991
dwBitErrorRate: 0
AvgTimePerFrame: 417082

BITMAPINFOHEADER:
biSize: 40
biWidth: 1280
biHeight: 688
biPlanes: 3
biBitCount: 12
biCompression: RV40
biSizeImage: 1320960
biXPelsPerMeter: 0
biYPelsPerMeter: 0
biClrUsed: 0
biClrImportant: 0

pbFormat:
0000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
0010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
0020: 5f 8f 08 00 00 00 00 00 3a 5d 06 00 00 00 00 00 _......:]......
0030: 28 00 00 00 00 05 00 00 b0 02 00 00 03 00 0c 00 (..............
0040: 52 56 34 30 00 28 14 00 00 00 00 00 00 00 00 00 RV40.(..........
0050: 00 00 00 00 00 00 00 00|00 00 00 22 56 49 44 4f ..........."VIDO
0060: 52 56 34 30 05 00 02 b0 00 0c 00 00 00 00 00 17 RV40...........
0070: f9 de 01 08 10 20 40 00 60 00 ... @.`.



But the video clip is 720P that to large to share samples...

[tetsuo55]

I have not tested HD rmvb, only SD,

It might be related to HD only?

[nijiko]

I have not tested HD rmvb, only SD,

It might be related to HD only?

Mmm. I think it also probably related to Real offical decoder.
Just probably.
Because I installed the RealPlayer11 (11.1.0 / 6.0.14.895).

[tetsuo55]

Mmm. I think it also probably related to Real offical decoder.
Just probably.
Because I installed the RealPlayer11 (11.1.0 / 6.0.14.895).

i believe realplayer 11 is (partly) incompatible with MPC-HC

[psme]

Great work! But there is heavy tearing at 24p output, both windowed and full screen. EVR or Overlay works fine without tearing. At 60Hz output, madVR don't tear, but I must play 24p source at 24p output!

My system, E6600 C2D at 3G, Win XP SP2, ATI 4850 512M (core 250Mhz, memory 750Mhz, underclocked but that has no effect on tearing), driver Cat 8.10 I think.

Attached a madVR OSD data.

When it can run smooth in 24p, I'll give it a serious try! ;)

regards,

Li On

[ice25]

Great work! But there is heavy tearing at 24p output, both windowed and full screen. EVR or Overlay works fine without tearing. At 60Hz output, madVR don't tear, but I must play 24p source at 24p output!

My system, E6600 C2D at 3G, Win XP SP2, ATI 4850 512M (core 250Mhz, memory 750Mhz, underclocked but that has no effect on tearing), driver Cat 8.10 I think.

Attached a madVR OSD data.

When it can run smooth in 24p, I'll give it a serious try! ;)

regards,

Li On

Yup i confirm, had a quick test at 24hz and indeed massive tearing, no prob at 48/50/60.

[mark0077]

I don't think madshi has gotten to adding vsync code yet so even 24fps at 24hz therefore won't always be tear free right? If one of those frames is "half" drawn when your screen does one of its redraws, then youll see half a frame / aka tearing.

This is my understanding of it anyways. Its like if I play an old game and set its max fps to 60, and my display to 60hz, I will still see tearing because thos 60 frames coming from the game arn't synced with the 60 refreshes of my display.....

Correct me if I am wrong but this is why i think your seeing tearing... we need to be patient :D This renderer is going to be :O:O as time goes on.

[yesgrey]

I was being partly sarcastic.
Sorry, i've missed it.

Great work! But there is heavy tearing at 24p output, both windowed and full screen.
Have you tryed setting your VSync in the drivers always On? Maybe it could help...

[psme]

Have you tryed setting your VSync in the drivers always On? Maybe it could help...

Tried, same tearing. At 60Hz, no tear with or without vsync.

regards,

Li On

[madshi]

w/o a perfect display and a $2000 probe(freshly recalibrated please) in a pitch black room w/ black painted walls, you won't get any benefit from all this...
Either you have a probe which delivers reasonably good results or you don't. If you have such a probe, having a complete CMS with corrections at multiple stimulus levels should give a noticeably better result than doing just a simple correction. If you don't have a reasonably exact probe, then you're screwed either way.

If you have a perfect display then you don't need any CMS, neither simple nor complex.

Obviously I don't agree with your argumentation at all.

But there is heavy tearing at 24p output, both windowed and full screen. EVR or Overlay works fine without tearing. At 60Hz output, madVR don't tear, but I must play 24p source at 24p output!
Currently I only have a 60Hz LCD computer monitor, so I can't really test anything other than 60Hz. But I've a Sony HW10 on order which will hopefully appear soon. Then I can test all the funny stuff like 24Hz with primary and secondary graphics port etc...

[leeperry]

Either you have a probe which delivers reasonably good results or you don't. If you have such a probe, having a complete CMS with corrections at multiple stimulus levels should give a noticeably better result than doing just a simple correction. If you don't have a reasonably exact probe, then you're screwed either way.

If you have a perfect display then you don't need any CMS, neither simple nor complex.
what I meant by "perfect" was "with spot-on primaries/secondaries saturations"...which Colorfacts won't let you measure.

so even if you had a top-notch Minolta spectrophotometer, you wouldn't be able to measure the primaries/secondaries saturations.

that was my point....only Color.HCFR does that, and the only high-end sensor it supports is the $2000 Eye One Pro...but sure, it cannot hurt! the more accurate the better :cool:

all I was saying is that a $200 sensor is not accurate enough to witness any improvement(99% of users who will use this?)...a ΔE of 1 can mean 3/6/8 or 10 w/ an i1pro/CS200..

anyway your point is that you want it more perfect than perfect, sounds good to me! :thanks:

[madshi]

@leeperry, does it make sense to do home calibration with a $200 sensor at all? I have my doubts if those are good enough. But I have to admit that I'm not really an expert in which sensors can do what for which price...

[leeperry]

well, it depends on your display actually...front projection w/ LCD or DLP : yes...SXRD : mixed bag IIRC...the rest I don't really remember, Plasma can be a problem I think but LCD is cool.

but most of the expensive units(like the i1pro) need to have their black level internally recalibrated every 15 mins(or less), so you literraly spend your life putting the damn thing in a bag and click "AUTO-CALIBRATION" :D

here's some links on AVS where Tom Huffman compares the $200 i1d2 against the $2000 i1pro :

http://www.avsforum.com/avs-vb/showthread.php?p=9495885#post9495885

http://www.avsforum.com/avs-vb/showthread.php?p=9503169#post9503169

too bad the pictures are down, because they were self-explanatory

All in all, I can't recommend the Spyder2. The Display2, especially the LT version, is very inexpensive and performs much better. In fact, at least with a DLP projector it performs essentially the same as the much more expensive EyeOne Pro spectroradiometer.
Even more surprising is how close the Display2 and Eye One Pro spectroradiometer are in performance. Between 20-100 IRE the Display2 reads slightly less red at 100 IRE, but I doubt that the difference between the calibrations would be visible. With respect to chromaticity, the area of traditional strength of spectroradiometers, there is a small difference in green and blue. Again, I seriously doubt whether these differences would be visible.

This confirms my earlier assessment that this is a very impressive inexpensive meter

that's my point, you can nitpick night and day about colorimetry, but you quickly reach a level of invisible stuff...even Mr Huffman said it might be overkill initially(and changed his mind later on you told me).

you don't plan on recalibrating your display every week, right? well mastering studios do...and the tools that are used to do this are internally factory recalibrated every few months.

but that stimulus stuff will be done via cr3dlut I guess?

[cyberbeing]

Leeperry, the i1pro isn't a $2000 part, it's more like a $500 part with very overpriced software. Your point still stands of course, considering the i1Display is a $100 part with somewhat reasonable software. If you get lucky you may be able to find a i1pro at a good price. Last Christmas, I bought my brand new i1pro retail bundle for $537 via Amazon.com.

As Leeperry mentioned though, if you have a display that is compatible with the i1Display it does a perfectly respectable job on those limited display types and you would only see a very slight improvement with the i1Pro.

[bur]

The screenshots in the first posting look like there was a huge difference between ffdshow and haali/madvr. I can't see any difference at all (maybe haali is a little less saturated) on my system (8800GT, Win XP) even when comparing on a per pixel level.

I often read ffdshow RGB conversion was supposed to be bad, but I absolutely can't confirm that.

I took the shots with MPC HC and ffdshow 2925. I anyone's interested I could upload some. But I'd also like to know if other's really get a picture that much worse with ffdshow. Or that much better when using madvr.

[leeperry]

Leeperry, the i1pro isn't a $2000 part, it's more like a $500 part with very overpriced software. Your point still stands of course, considering the i1Display is a $100 part with somewhat reasonable software. If you get lucky you may be able to find a i1pro at a good price. Last Christmas, I bought my brand new i1pro retail bundle for $537 via Amazon.com.

As Leeperry mentioned though, if you have a display that is compatible with the i1Display it does a perfectly respectable job on those limited display types and you would only see a very slight improvement with the i1Pro.
well the i1d2 is an improved i1lt and costs $200 on ebay and most US online shops : http://www.amazon.com/Pantone-MEU103-Eye-One-Display-2/dp/B000CR78CO

and the i1pro costs at least a grand in europe, guess you got really lucky!

manufacturers software is too dumbed down to my taste...ArgyllCMS & Color.HCFR are perfect! ArgyllCMS will take 20 mins to build a LUT file, and when you double-check it in Color.HCFR it's spot-on :)
I'd also like to know if other's really get a picture that much worse with ffdshow. Or that much better when using madvr.
it mostly shows w/ test patterns and red objects, try the start credits of SHOOT'EM UP, with the leaking blood letters.

[Mark_A_W]

@leeperry, does it make sense to do home calibration with a $200 sensor at all? I have my doubts if those are good enough. But I have to admit that I'm not really an expert in which sensors can do what for which price...

Madshi, my friend is a professional calibrator with Colorfacts, and we compared HCFR with the DIY HCFR probe to Colorfacts and got basically identical results.

We also used Colorfacts to generate a calibration file for my projector (HFCR needs that), but even without that calibration, HCFR was very, very close.

The answer, at least in the case of HCFR, is yes it most certainly is good enough (maybe we should have a whip around and try and get you one? Leeperry?).


Also, do you have a CRT monitor Madshi? A CRT monitor will not mess with the timings of the video signal at all. It's a cheap and easy way to get you a second display that will work with almost all refresh rates (but not 24hz), and as it's true multisync, what you see is what you get.

These days you just grab the fanciest one you can find dumped on the side of the road ;)

They are perfect for testing tearing, as it cannot cause tearing itself. Well, except at 24hz, but it's a start.


Thanks

Mark

[cyberbeing]

well the i1d2 is an improved i1lt
The only difference between the i1d2 and the i1lt is the software. The hardware puck is absolutely identical. With that said, if you don't plan to use the Eye One Match software, buying the i1d2 is a waste of money.

[leeperry]

The only difference between the i1d2 and the i1lt is the software. The hardware puck is absolutely identical. With that said, if you don't plan to use the Eye One Match software, buying the i1d2 is a waste of money.
it might be, but apparently the firmware has been improved...and anyway, the LT is really old now and god knows how these sensors will do after a couple of years.

even if you have to auto-calibrate the device before each calibration session, obviously it might not "age" too well..there's no available data about this actually.
We also used Colorfacts to generate a calibration file for my projector (HFCR needs that), but even without that calibration, HCFR was very, very close.

The answer, at least in the case of HCFR, is yes it most certainly is good enough
the HCFR sensor is too dependent on reference files IMO, which are also dependend on many other things...that's its biggest flaw.

w/o a reference file, you can't reaxh D65 because there's no white reference, you can only measure gamma and gamut...and if it doesn't exactly match your unit, it will give inaccurate results.

also, it will only work for front calibration...and the spyder3 has big problems w/ wide gamuts apparently(the spyder2 being worthless).

the i1d2 still has the crown, it's rebadged as "lacie blue eye pro"(the one I got) and it's definitely the best bang for the bucks!

[Thunderbolt8]

some questions on input&output leves, I can set them in ffdshow and also in your renderer (at least the output ones). so where to select which now? I can choose between auto, standard, full range in ffdshow for input and between computer monitor and tv/projector for output there. and in addition, I can also set output levels in your renderer now. so whats now the recommended way of settings in those 3 areas for 1) HD stuff and 2) SD stuff ?(Im watching on an LCD TV display and not on a pc screen)

[FoLLgoTT]

The difference between $200 i1Display2 (colorimeter) and $800 Eye One Pro (spectrophotometer) is not very big in real world use. The difference depends on the display technology measured (CRT/LCD etc.). Usually the primarys and grey levels are more or less a bit off with the Display2. It is not temperature compensated and it measures repeatable down to 10IRE. The I1Pro has to be calibrated every few minutes and is useless under 30 IRE. Only with CALMan's statistical analysis it gives repeatable results down to about 10 IRE.

Since every sensor has its measurement errors to question the effort to perfect CMS accuracy seems reasonable. I agree with leeperry when he states that for real world use 100IRE gamut correction is enough.

[Mark_A_W]

the HCFR sensor is too dependent on reference files IMO, which are also dependend on many other things...that's its biggest flaw.

w/o a reference file, you can't reaxh D65 because there's no white reference, you can only measure gamma and gamut...and if it doesn't exactly match your unit, it will give inaccurate results.
!

Like I said, HCFR without a calibration file was almost EXACTLY the same as Colorfacts, at least on a CRT projector.

Calibrated against Colorfacts, it was obviously identical.

Also, in the real world, minor differences are not visible to the eye. There is no way I could pick the difference between the calibration using HCFR without a reference file, and a calibration with the reference file, and I have perfect colour vision.

You worry too much about such things.

[leeperry]

Like I said, HCFR without a calibration file was almost EXACTLY the same as Colorfacts, at least on a CRT projector.
well if you don't have a reference file, you have to build one..w/ a better sensor(i1d2/i1pro) ideally, but if you don't have one you have to learn Color.HCFR what your destination white point is by making it w/ the HCFR unit...so you have to "guess" what D65 white looks like.

I've had a HCFR sensor and it won't work w/o a ref file, I even discussed it thoroughly w/ the main HCFR color engineer.

honestly the HCFR unit is worthless, I'd rather even have a spyder2..

[Mark_A_W]

honestly the HCFR unit is worthless, I'd rather even have a spyder2..

It worked fine without a reference file - you can generate a quick reference file to get you going.

The result of this was close enough to Colorfacts for me.

Was it ideal? No.

Was it good enough? Yes.

[nijiko]

i believe realplayer 11 is (partly) incompatible with MPC-HC

But in VMR9 and other such renders, the film works fine.
So I surely consider the problem is about madVR.
Or the compatible about madVR.

[tetsuo55]

But in VMR9 and other such renders, the film works fine.
So I surely consider the problem is about madVR.
Or the compatible about madVR.

Strange,

Can you try uninstalling real11 and install realalternative? (www.codecguide.com)

[kostik]

Is there anything to do in order to watch movies with external subs with this renderer?

[nijiko]

Strange,

Can you try uninstalling real11 and install realalternative? (www.codecguide.com)

Oh yeah! Crash again!
This time the same crash with AVI files~~
And this time is each 30min~~
Video and audio decoder are same ffdshow.
However, no problem with other renders.

This AVI:

滤镜 : ffdshow Video Decoder - CLSID : {04FE9017-F873-410E-871E-AB91661A4EF7}

- 连接到:

CLSID: {1B544C20-FD0B-11CE-8C63-00AA0044B51E}
滤镜: AVI Splitter
Pin: Stream 00

- 连接的媒体类型:

Video: DivX 5 640x480 29.97fps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Video {73646976-0000-0010-8000-00AA00389B71}
subtype: Unknown GUID Name {30355844-0000-0010-8000-00AA00389B71}
formattype: FORMAT_VideoInfo {05589F80-C356-11CE-BF01-00AA0055595A}
bFixedSizeSamples: 0
bTemporalCompression: 1
lSampleSize: 1
cbFormat: 88

VIDEOINFOHEADER:
rcSource: (0,0)-(0,0)
rcTarget: (0,0)-(0,0)
dwBitRate: 0
dwBitErrorRate: 0
AvgTimePerFrame: 333666

BITMAPINFOHEADER:
biSize: 40
biWidth: 640
biHeight: 480
biPlanes: 1
biBitCount: 24
biCompression: DX50
biSizeImage: 921600
biXPelsPerMeter: 0
biYPelsPerMeter: 0
biClrUsed: 0
biClrImportant: 0

pbFormat:
0000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
0010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
0020: 00 00 00 00 00 00 00 00 62 17 05 00 00 00 00 00 ........b.......
0030: 28 00 00 00 80 02 00 00 e0 01 00 00 01 00 18 00 (...€..........
0040: 44 58 35 30 00 10 0e 00 00 00 00 00 00 00 00 00 DX50............
0050: 00 00 00 00 00 00 00 00 ........



滤镜 : ffdshow Audio Decoder - CLSID : {0F40E1E5-4F79-4988-B1A9-CC98794E6B55}

- 连接到:

CLSID: {18C16B08-6497-420E-AD14-22D21C2CEAB7}
滤镜: 音频切换器
Pin: [PV][Mizuki Nana] NANA CLIPS.avi / Stream 01

- 连接的媒体类型:

音频: PCM 48000Hz 立体声 1536Kbps

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Audio {73647561-0000-0010-8000-00AA00389B71}
subtype: MEDIASUBTYPE_PCM {00000001-0000-0010-8000-00AA00389B71}
formattype: FORMAT_WaveFormatEx {05589F81-C356-11CE-BF01-00AA0055595A}
bFixedSizeSamples: 1
bTemporalCompression: 0
lSampleSize: 1
cbFormat: 18

WAVEFORMATEX:
wFormatTag: 0x0001
nChannels: 2
nSamplesPerSec: 48000
nAvgBytesPerSec: 192000
nBlockAlign: 4
wBitsPerSample: 16
cbSize: 0 (扩展字节)

pbFormat:
0000: 01 00 02 00 80 bb 00 00 00 ee 02 00 04 00 10 00 ....€.........
0010: 00 00 ..

- 已枚举的媒体类型 0:

设置为当前的媒体类型

- 已枚举的媒体类型 1:

未知

AM_MEDIA_TYPE:
majortype: MEDIATYPE_Stream {E436EB83-524F-11CE-9F53-0020AF0BA770}
subtype: MEDIASUBTYPE_None {E436EB8E-524F-11CE-9F53-0020AF0BA770}
formattype: FORMAT_None {0F6417D6-C318-11D0-A43F-00A0C9223196}
bFixedSizeSamples: 0
bTemporalCompression: 0
lSampleSize: 307200
cbFormat: 0

[Hypernova]

Is there anything to do in order to watch movies with external subs with this renderer?

Use something like ffdshow and enable subtitle. Or, like me, wait. Madshi said he will add it, but on what priority I don't know. (May I ask, Mashi?) After smoothness, subtitle is the feature that I'm waiting for.

[tetsuo55]

Oh yeah! Crash again!I give up, i cannot reproduce on my system.

Windows7 7100 x86
Hd2400pro CAT9.5 beta

You should create a minidump

[madshi]

The screenshots in the first posting look like there was a huge difference between ffdshow and haali/madvr. I can't see any difference at all (maybe haali is a little less saturated) on my system (8800GT, Win XP) even when comparing on a per pixel level.
Chroma upsampling differences are sometimes very evident and sometimes barely (if at all) noticeable. It all depends on the material you're testing with. Usually the best material to see differences is with "red-on-black" fonts like shown in the first post of this thread. So find yourself a movie with a scene where there is lots of red colored stuff on a black background. You should be able to see differences there. E.g. one scene I stumbled over recently was where Sam gets connected to the Cylon Baseship the first time (Battlestar Galactica). That scene is all red and black. The scene was totally unwatchable with VMR9 on my PC, while it looked just fine to me with madVR.

Madshi said he will add it, but on what priority I don't know. (May I ask, Mashi?)
Priorities right now are smooth playback, DVD playback and after those two probably subtitles.

Oh yeah! Crash again!
This time the same crash with AVI files~~
And this time is each 30min~~
Are all the crashes you report after many minutes? Or are some of them at once when you start playback? Those crashes after many minutes have never been reproduced by anyone else yet, as far as I know. I don't know where those crashes are coming from, but since nobody else seems to have those, I'd guess maybe it's a bad graphics driver? Try updating your graphics drivers. And maybe also your GPU's BIOS, if there is a newer one available.

what I meant by "perfect" was "with spot-on primaries/secondaries saturations"...which Colorfacts won't let you measure.

so even if you had a top-notch Minolta spectrophotometer, you wouldn't be able to measure the primaries/secondaries saturations.

that was my point....only Color.HCFR does that, and the only high-end sensor it supports is the $2000 Eye One Pro...but sure, it cannot hurt! the more accurate the better :cool:

all I was saying is that a $200 sensor is not accurate enough to witness any improvement(99% of users who will use this?)...a ΔE of 1 can mean 3/6/8 or 10 w/ an i1pro/CS200..
This all confuses me a bit. On the one hand you're saying that the i1 D2 is nearly identical in performance to the i1 Pro (at least with specific types of displays). On the other hand you're saying that no $200 sensor is accurate enough to bring improvements. Is it just me, or does that contradict each other?

BTW, I've asked TomHuffman via PM what a perfect CMS would have to do in his opinion and he replied that he would like to have different corrections at 25% step stimulus levels. And even different corrections at 25% step saturation levels on top of that!

The answer, at least in the case of HCFR, is yes it most certainly is good enough (maybe we should have a whip around and try and get you one? Leeperry?).

Also, do you have a CRT monitor Madshi? A CRT monitor will not mess with the timings of the video signal at all. It's a cheap and easy way to get you a second display that will work with almost all refresh rates (but not 24hz), and as it's true multisync, what you see is what you get.
I don't have space for a CRT. But the Sony HW10 I'll get soon enough should do all important refresh rates.

I might get a spectrometer sooner or later, but the CMS/3dlut work is done by yesgrey3, anyway, so it's not really that important for me to have one.

My general point of view about calibration is that sooner or later we will have laser light sources. Those should not change as much (if at all) over lifetime. Which means that with such light sources it might make sense to hire a professional calibrator and calibrate your projector once and for all with the best spectrometer available out there. And if you do that, a perfect CMS should bring benefits over a simple one. With current light sources things are different cause basically to get perfect results you'd have to recalibrate every couple of lamp hours. Obviously hiring an ISF guy every couple of lamp hours doesn't make any sense. So right now a perfect CMS might not bring that much benefits. But I don't like my work to be dictated by limitations of today's inferior hardware.

[tetsuo55]

nijiko, your videocard might be overheating?

[FoLLgoTT]

But I don't like my work to be dictated by limitations of today's inferior hardware.

Not only the hardware is limited, but the whole concept of colorimetry. The spectral curves of the CIE standard observer from 1931 are an average of very few persons. They were even extrapolated from different sources and filtered. So there may be visible error from person to person.

In terms of color temperature and gamut there is just no right or wrong, since everything is based on this inaccurate standard observer. Even the better newer observers (e.g. 1964) are just based on statistics. Colorimetry can never be an exact science, because colors exist only in our heads and don't exist in physics. Funny enough that it works good enough for real world use. ;)

[madshi]

some questions on input&output leves, I can set them in ffdshow and also in your renderer (at least the output ones). so where to select which now?
The levels controls in ffdshow only have any effect if ffdshow outputs RGB. madVR does not support RGB input, it only supports YV12 input. So there's no way you can make use of the ffdshow levels controls when using madVR, because if ffdshow outputs YV12, the ffdshow levels controls have no effect whatsoever.

[leeperry]

Colorimetry can never be an exact science, because colors exist only in our heads and don't exist in physics. Funny enough that it works good enough for real world use. ;)
well put! as long as our brain will be tricked by this kind of things, don't expect to see any improvement up to a certain level : http://www.chessbase.com/puzzle/puzz10b.htm
It worked fine without a reference file - you can generate a quick reference file to get you going.

The result of this was close enough to Colorfacts for me.
well I cannot rely on my eyes to make a D65 white reference file...great if you can.

[madshi]

In terms of color temperature and gamut there is just no right or wrong [...] colors exist only in our heads and don't exist in physics.
Then why calibrate at all? :rolleyes:

[FoLLgoTT]

Then why calibrate at all? :rolleyes:

This is a good question. I think the standard observer is sufficient for most cases. Our eye is not as precise as a spectrophotometer and it strongly depends on environment and many other factors. I think it is sufficient to calibrate the necessary values into a small range which is small enough that no difference can be seen with a before/after comparison with my own eyes. Differences which I can only see in a direct comparison are not important, because I never do this when watching movies (nobody watches at two screens simultaneiously ;)).

I just think practical. I think to have a very flexible CMS is very nice, but usually you only need the basics (including simple gamut correction) to get good results.

[leeperry]

Then why calibrate at all? :rolleyes:
all that is needed to trick our brain is proper grayscale, D65 white point, ±2.3 gamma and 100 IRE gamut conversion.

doing a D65 calibration gives proper grayscales, so colors are not "mixed up" anymore...and 100 IRE gamut conversion brings them back in the "visible" scale where you can enjoy them and think to yourself "now, that looks good!"

proper gamut conversion seems to increase the contrast ratio, because colors are as genuine as your brain could possibly see them.

look at the feedback from this guy on his JVC RS1/HD1 :
http://www.avsforum.com/avs-vb/showpost.php?p=11937064&postcount=90

It's a huge improvement in greens and reds, no more sunburnt faces.

it could prolly be mathematically improved by that stimulus stuff, but would it be visible? maybe! but if he doesn't recalibrate his display on a weekly basis, this won't matter anymore..