Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Hendrik]

JRVR, or JRiver Video Renderer, was created in MC28 and further developed in MC29. It has matured into a proper high-quality video renderer which "just works".

In Media Center 30 there is still more already and more to come!

MC30 Improvements so far
- Built-in deinterlacing independent of the GPU driver (and on all platforms!)
- Performance Presets for an easier setup
- More flexibile scaling options
- Reliability/stability improvements

Planned
- (Maybe) Dolby Vision Enhancement Layer support (for UltraHD Blu-ray discs) (under investigation)

General Video improvements
- Video Analysis, similar to the well known Audio Analysis
  - Cropping information for semi-automated blackbar cropping
  - More metadata in the library (HDR info, and more)
- Tonemapping thumbnails for HDR videos
- HDR -> SDR tonemapping for video conversion and streaming
- (Maybe) HDR video streaming

These are the planned changes, and will be developed during the lifetime of Media Center 30. Not every enhancement named here will be available right away.

What happened in MC29?
You can check what we achieved in MC29 for JRVR here:
https://yabb.jriver.com/interact/index.php/topic,132317.0.html

[jmone]

Great List!

[HTPC Videophile]

- (Maybe) Dolby Vision Enhancement Layer support (for UltraHD Blu-ray discs)

General Video improvements
- Video Analysis, similar to the well known Audio Analysis
  - Cropping information for semi-automated blackbar cropping
  - More metadata in the library (HDR info, and more)

(1)Is the Dolby Vision Enhancement layer support  for m2ts rips/remuxes from UHD discs or only MKV remuxes of UHD Bds ?

(2)For video Analysis, please ensure that the side-car files are not stored in the same library where the video files are stored(as was the unfortunate case in audio-analysis). They should be stored in a separate folder with links to the main library files.

[tkolsto]

"Built-in deinterlacing independent of the GPU driver"

I read about deinterlacing. This is to get better perceived motion and reduce flicker. So this is going to done seperately away from the GPU and handled within JRiver? Did I understand this correctly? And I interpret this to be a better way of rendering video?

Looking forward to this greatly! Maybe my problems will be less troublesome. Also the presets is a welcomed feature

[Hendrik]

(2)For video Analysis, please ensure that the side-car files are not stored in the same library where the video files are stored(as was the unfortunate case in audio-analysis). They should be stored in a separate folder with links to the main library files.

The same sidecar will be used, its not a "audio analysis sidecar" file, its generic. You can turn off sidecars and just use the MC library to store the data if you don't want it there.
The entire idea of the sidecar file is to have it next to the video, which means as long as the video file exists, its likely for the sidecar to go with it - if it were moved to a central location, that feature would be lost, and all you get is basically the MC Database, which already stores all this in a central location.

I personally turn the sidecars off as well, and just make sure my MC Database has regular backups as needed. Thats surely an option you can go for if you don't want the sidecar next to your file.

[Hendrik]

"Built-in deinterlacing independent of the GPU driver"

I read about deinterlacing. This is to get better perceived motion and reduce flicker. So this is going to done seperately away from the GPU and handled within JRiver? Did I understand this correctly? And I interpret this to be a better way of rendering video?

Looking forward to this greatly! Maybe my problems will be less troublesome. Also the presets is a welcomed feature

I'm not sure you quite understand what interlacing and deinterlacing is. Content can be interlaced, which is an inherent property of the video - typical for live TV content (prominently sports, concerts, and such events). Our displays are not interlaced, so we need to convert them - enter deinterlacing.

This planned change has no advantages beyond dealing with videos of such a nature.

[Smack]

Hello Hendrik

I just saw that in the upcoming mc30 release the black bar detection might be implemented.

Just out of curiosity how does it work. Do I have eggheads (like the conheads) because the picture has to be streched? Or will there be left and right some of the picture cropped?

[Hendrik]

It will not cut off any image (unless you want it to), nor stretch the image in unnatural ways.

People with an ordinary 16:9 screen won't really make use of this, its useful for wider screens and projectors.

[HTPC Videophile]

The same sidecar will be used, its not a "audio analysis sidecar" file, its generic. You can turn off sidecars and just use the MC library to store the data if you don't want it there.
The entire idea of the sidecar file is to have it next to the video, which means as long as the video file exists, its likely for the sidecar to go with it - if it were moved to a central location, that feature would be lost, and all you get is basically the MC Database, which already stores all this in a central location.

I personally turn the sidecars off as well, and just make sure my MC Database has regular backups as needed. Thats surely an option you can go for if you don't want the sidecar next to your file.

  thanks for the confirmation Hendrik regarding this. One adjunct query regarding the same: For the  case  of varying width of black bars within the same movie, will the video analysis take care of it(eagerly expected scenario for me !) or will it resolve to some nominal value that just removes/minimizes the black bars (Unwanted/Unacceptable scenario for me !). And if i don't mind stretching/distorting my videos to fill the screen , can i stretch the cropped videos(using video analysis info) to full screen(16:9) display like i do after MadVR automatic cropping.Will it give the same results as far as final image co-ordinates are concerned(compared to MadVR) . if my query is NOt clear i might post screen shots of video image with MadVr cropping/processing and  then MC  stretching .     Also please  clarify the query(1) in my above post .

[Hendrik]

The first version will most likely not support varying aspect ratios. Its a niche feature used in only a very small minority of movies, and often also part of the artistic intent.

The entire idea is to not detect blackbars live during playback, because thats computationally rather expensive and complex, which is why madVR for example only allows it with CPU decoding and not when hardware accelerated decoding is used (and its slow).
Instead, the black bars will be detect beforehand, and the information only used during playback.

[jgreen]

Yes, presets! 
For "Drama", I'm sure you've got it all well-covered, but do not forget "Sports".  For action sports in particular, motion-smoothing becomes more important.   The action is typically smaller-scale and confined to the center of the image. 

And a feature request:  This should be VERY easy, and yet no one is doing it.  Can we have a zoom function for video?  Say full screen is 100%, can we zoom to 150% or even 200%?  These days sports programming is infested with scroll banners at the bottom and sides.  Meanwhile the actual sports action has gotten smaller and is still center-image, easily zoomable with today's superb algos.   And while we're at it, why not a mouse-actuated pan feature?  With that I could do a slight zoom and slide the blessed banners offscreen.  Goodbye to important messages and programming notes! 

[Hendrik]

Zoom and Pan should already be supported. On the mouse, hold Shift, and then zoom with mouse wheel and pan by just click and drag. You can also do it through the OSD with the arrow keys or a remote control, although with the mouse is easier.

PS:
Note that presets are currently performance based. But we already support user-defined profiles you can use to setup for different types of content, like sports.

[jgreen]

Er, I'm running 29.X.86.  That version and all previous have a quite rudimentary "100%/200%/Full Screen" option, but still no mouse control or pan--mouse scroll moves the video fwd or back.

If this is already in v30.x, hooray.  If it's already in 29, could someone kindly deconfoodle me?

[Hendrik]

Hold down Shift, then it should work.

[jgreen]

Oh my, that was easy!

[cgott42]

I currently own MC29 and my sole usage is to better handle the upscaling and motion of 720p video (all network TV sports) on a 4K OLED TV
For this purpose - what benefits could I expect in MC30 (I'm asking to save the trouble of installing the MC30 demo)

TIA

[JimH]

No MC30 demo is available yet.

https://yabb.jriver.com/interact/index.php/topic,133708.0.html

[cgott42]

Gotcha - what benefits could I expect in MC30 re: the upscaling and motion of 720p video (all network TV sports) on a 4K TV

[armyplace]

- HDR -> SDR tonemapping for video conversion and streaming

Can you tell us more about this feature, are you talking about streaming services or something else?

Any improvements in JRVR decoding performance on older gfx cards, eg nvidia GTX1650/1070

thanks

[datdude]

I think steaming is referring to MC’s internal streaming engine to Panel, JRemote, etc.

[Hendrik]

I think steaming is referring to MC’s internal streaming engine to Panel, JRemote, etc.

Indeed, quite correct. Currently you get wrong colors if you were to stream a HDR movie to JRemote

[masterful]

would it be possible to support these two features in MC30:

- popup thumbnails like we see in YouTube's player ? this feature is very handy when trying to jump to points of interest in a video.
- an ability to add bookmarks to video files manually. when they are set we could have a thumbnail of the scene that was bookmarked so that we can easily identify a scene of interest from many bookmarked scenes.

[jachin99]

Will video analysis only work with local files or will it also work with URLs like those sent from the hd homerun tuners

[Hendrik]

Will video analysis only work with local files or will it also work with URLs like those sent from the hd homerun tuners

Can't analyse a live stream or such, no.

[yaddo]

- (Maybe) Dolby Vision Enhancement Layer support (for UltraHD Blu-ray discs)

Wow, this is huge. This will theoretically fix those discs with a badly encoded base layer which gets "patched" by the Enhancement Layer (many Studio Canal titles for example), right?

[HTPC Videophile]

- (Maybe) Dolby Vision Enhancement Layer support (for UltraHD Blu-ray discs)

  - More metadata in the library (HDR info, and more)

(1)Does the Dolby Vision Enhancement Layer support (for UltraHD Blu-ray discs) include support for UHD m2ts rips/remuxes from the UHDs as well(say using DVD-Fab UHD ripper)? I am specifically asking this because earlier there was mention for limited support for mkv rips only for some Dolby Vision capabilities of JRVR although container format should not make any difference.
(2)With more metadata in the library, can we take the indulgence to expect new HDR info field  to be displayed in  the Standard and Theater View(mentioning Dolby Vision/HDR10+/HDR10 ) just like FPS, Bitrate, Compression etc fields currently? Also, why can't we integrate SOT(Swag of Tools) relevant code/capability using mediainfo to add the Dolby Atmos/DTS-X info in MC itself instead of using the third party tool . As a humble suggestion, the HDR info and additional audio info should be accomodated in the current  Compression field label  itself justthe way  like SOT tool does (for example:m2ts video(video: hevc hdr10+, audio:truehd + Dolby Atmos)

[jmone]

(2)With more metadata in the library, can we take the indulgence to expect new HDR info field  to be displayed in  the Standard and Theater View(mentioning Dolby Vision/HDR10+/HDR10 ) just like FPS, Bitrate, Compression etc fields currently? Also, why can't we integrate SOT(Swag of Tools) relevant code/capability using mediainfo to add the Dolby Atmos/DTS-X info in MC itself instead of using the third party tool . As a humble suggestion, the HDR info and additional audio info should be accomodated in the current  Compression field label  itself justthe way  like SOT tool does (for example:m2ts video(video: hevc hdr10+, audio:truehd + Dolby Atmos)

Hendrik has already said he plans to expand the MetaData collection for Video in MC30, so much of what my SOT collects may become redundant at some point.  MC (as well as all other MediaPlayers) relies on FFMPEG (well the ffprobe subset), and hence, is limited to what by what it can report on.  The code to detect if an Audio Stream has Meta Data Extensions (ATMOS, DTS-X etc) does not exist in ffprobe (hence SOT using MediaInfo for this).  The "best" solution would be the addition of this capability in ffmpeg but that is well outside my expertise and unless someone is keen to add it..... it will not happen.  I do note that MediaInfo is also open sourced so all the logic to do this is already in the public domain. 

[bogdanbz]

I vote for:
- Dolby Vision Enhancement Layer support (for UltraHD Blu-ray discs)
- scaler improvements

Regarding scaler improvements, I am using Jinc as the luma scaler and I find that it has more visible aliasing than madVR's Jinc scaler. This is most easy to notice on anime, where the black thin contours around characters have aliasing steps on oblique lines on JRVR but not on madVR.

[bogdanbz]

There's another thing that might warrant some investigation regarding HDR.
- if I keep Windows HDR turned off, and use madVR to play HDR video with a nVidia graphic card, then madVR uses the NvAPI to switch the display into HDR10 mode, and to pass HDR static metadata to it
- if I keep Windows HDR turned on (as I have disabled the HDR auto-switch in JRVR) and use JRVR instead, then the same HDR10 video has different luminance and saturation than in the madVR case above

In this second case, I suspect metadata set on the swapchain with SetHDRMetaData is not sent to the display itself, but is used by the Windows DWM instead to perform a color space conversion of the video content. This CSC transform maybe converts the gamut to the one defined by the EDID data from the display, that usually does not have the same red, green, blue, white point chromacity coordinates as the BT2020 ones. The HDR luminance static metadata is also probably used at this point to tone map to a target of 1500 nits or something like that.

I suspect this is what happens because the VESA DisplayHDR utility, which VESA provides on Windows 10 & 11 systems to perform measurements of a HDR display for VESA HDR compliance, seems to use SetHDRMetaData to set as static metadata the values found in the EDID in order to avoid any processing by Windows itself. This can be seen in the Game.cpp file in the DisplayHDR source code, which is open source and available on GitHub.

[Hendrik]

In this second case, I suspect metadata set on the swapchain with SetHDRMetaData is not sent to the display itself, but is used by the Windows DWM instead to perform a color space conversion of the video content.

I have tested the transmitted metadata with a HDMI analyzer sitting between the display and the PC, and can state with certainty that the above is not the case, at least not in any generic capacity. Of course I cannot comment on your systems behavior.

Note that HDR pass-through is only accurate in fullscreen, so testing for example windowed mode may exhibit different results.

Please note that bug reports or other topics warranting further discussion should have their own thread.

[bogdanbz]

Nvidia has a setting in its driver, the "Reference mode", which is supposed to disable the use of the CSC or the Windows DCT: https://www.nvidia.com/content/Control-Panel-Help/vLatest/en-us/mergedProjects/nvdsp/CS_Adjust_Color_Settings_Advanced.htm

I don't have the nvAPI document available now, but the blocks which perform these operations on a nVidia graphic card are sitting between the  frontbuffer and the graphic card rasterizer. Among these blocks is the one that performs the gamma correction specified in a VCTG tag of an ICC profile too.

Although I have "reference mode" enabled in the driver, there's a noticeable difference in between the luminosity and saturation of HDR video content rendered by madVR when it uses the nvAPI to switch the display to HDR (in exclusive mode), and when using JRVR while Windows HDR was active.

This was my guess as to what is happening, I am not sure this is what happens. I just see the end result, the difference in the image on the TV connected to the graphic card in the two cases.

[Plutotype]

- (Maybe) Dolby Vision Enhancement Layer support (for UltraHD Blu-ray discs)

Hi Hendrik,

Is this please based on the DoVi_tool / DoViBaker? My understanding of these tools is that they enable to "pre-process" Dolby Vision FEL titles ( UHD discs ), which JRVR then would be able to read and render in HDR10 and send to the display? Please confirm.

Thanks
Pluto

[bogdanbz]

Feature suggestion: the use of Nvidia's FRUC library for video frame interpolation. It might be only for 4xxx series though. The Optical Flow unit is available on cards starting with Turing (2xxx series and more recent), but I don't know if this SDK can make use of it for video interpolation on them (they restrict the use of the unit to 4xxx series for DLSS, I don't know about video usage).

[eve]

Is there going to be a documented method for loading your own shaders into libplacebo?

shaders/custom.h: Allows directly ingesting custom GLSL logic into the pl_shader abstraction, either as bare GLSL or in mpv .hook format.

Lots of excellent mpv hook shaders out there that I'd love to be able to use.
 

[Hendrik]

We don't currently support loading your own custom shaders. Maybe something for the future.

[eve]

We don't currently support loading your own custom shaders. Maybe something for the future.

Thanks! <3 Would be a *huge* step towards replacing MadVR, allowing cross platform playback for critical viewing instead of being weirdly stuck with Windows (especially considering HDR is slowly working itself out on linux!!!).

JRVR is performant for say watching a movie while I work but doesn't really hold up for critical viewing when compared to MadVR. The differences are subtle but there. I've gotten MPV pretty close to MadVR using custom shaders so the introduction of user shader support would be huge (the groundwork has already been done in LibPlacebo too).

Thanks again Hendrik, your contributions to JRiver are incredible.

[eve]

Feature suggestion: the use of Nvidia's FRUC library for video frame interpolation. It might be only for 4xxx series though. The Optical Flow unit is available on cards starting with Turing (2xxx series and more recent), but I don't know if this SDK can make use of it for video interpolation on them (they restrict the use of the unit to 4xxx series for DLSS, I don't know about video usage).

The 4000 series interpolation might be quite good. Anything real time I've played with hasn't really passed the 'test' at not being distractingly.... artifact-y. You can tell it's interpolated, much like how TV interpolation gets quite ugly on fast 'illogical' motion.
Offline is a slightly different story. I have a test I've been using, the first light cycle scene from Tron Legacy taken from an open matted 1080p master that does look quite good. The thing is, even with nice hardware it's no where approaching real time and at the moment, the best way to cut down on strange artifacting is pre-processing the entire video to detect individual shots (or even segments of said shots). Motion interpolation works pretty well for CG heavy content with 'clean' live action plates IMO

[armyplace]

Does JRVR apply "dynamic" tone mapping or the only option is the plugin using madvr?

[tkolsto]

eve, when You talk about 4000 series..you mean rtx 4000 and that these new cards could be a lot better at working with interpolation and far fewer artifacts on not CG contant which is more demanding. Is this  what You mean? or is this some other kind of interpolation than a Tvs interpolation?

[eve]

eve, when You talk about 4000 series..you mean rtx 4000 and that these new cards could be a lot better at working with interpolation and far fewer artifacts on not CG contant which is more demanding. Is this  what You mean? or is this some other kind of interpolation than a Tvs interpolation?

Yeah. I'll have to play around with it when I get one (I'm probably just going with a pair of 3090s in the short term, less performance but more VRAM for the price and better for my AI / Rendering tasks).

With video, you won't have access to motion vectors or other 'metadata', but it's not unreasonable to think that some of the improvements won't be able to be applied to video OR that the increased processing capability wont allow us to get closer to real time perf using existing interpolation methods.
Rife (an offline interpolation method) is pretty solid but it requires handholding to get ideal results and extensive processing time. I use it for some HFR demos (including intercutting 24fps cinematography with HFR), and more extensive retiming of footage for work. It yields better results than the quick methods built into editing software but it's only worth it for situations where re-timing would be noticeable.

[thorsten]

Sorry  eve, but I think there are way more important features than some fancy stuff with a gpu that is high priced and not really available… 

[thorsten]

Oh, and I vote for the (offline is fine) black bar cropping!

[eve]

Sorry  eve, but I think there are way more important features than some fancy stuff with a gpu that is high priced and not really available… 

Oh I wasn't really suggesting it for JRVR. Though it may not be absurd to implement some form of interpolation down the line, especially as the methods get more 'pre-packaged' and easy to implement.
Plus, the topic of Motion Interpolation isn't exclusive to the 4000 series of GPUs. God knows we've been doing it for ages without them. Rife is still my primary go to. However, Topaz does a good job with older 59i and 60i sources at getting really smooth output. You can absolutely duplicate the correct 'soap opera' smoothness effect that, depending on your source, may be how it was intended.


I've already given my spiel about what JRVR needs

Custom shader support for users that know what they're doing
Full MCWS control of all video parameters
Dolby Vision with FEL (good luck  ) support.

[Hendrik]

In the next build for JRVR:

NEW: GPU compute accelerated shader deinterlacing in JRVR (based on the YADIF algorithm).

Using a custom shader implementation for deinterlacing should result in more consistent results. Quality comparisons to the video hardware deinterlacing through D3D11 is impractical in general, as it can vary based on your GPU and drivers.

The big bonus here is that it also offers GPU-accelerated deinterlacing on Linux/Mac, which haven't had any access to that until now.

You can select the deinterlacing mode to use in the JRVR settings under Processing.

[jmone]

The new DeInt option looks good to me.  Did a comparison with that sample clip and DeInt worked much better!
Thanks
Nathan

[eve]

In the next build for JRVR:

NEW: GPU compute accelerated shader deinterlacing in JRVR (based on the YADIF algorithm).

Using a custom shader implementation for deinterlacing should result in more consistent results. Quality comparisons to the video hardware deinterlacing through D3D11 is impractical in general, as it can vary based on your GPU and drivers.

The big bonus here is that it also offers GPU-accelerated deinterlacing on Linux/Mac, which haven't had any access to that until now.

You can select the deinterlacing mode to use in the JRVR settings under Processing.

Thank you! That's awesome.
Deinterlacing in real time is always somewhat meh but it's gotten way better over the years.
I've been actually dealing with a bunch of interlaced content for offline upscaling recently and it's been annoying at best. QTGMC seems to work best but it's understandably not super fast.

[tzr916]

NEW: GPU compute accelerated shader deinterlacing in JRVR (based on the YADIF algorithm).

When I enable JRVR Shader YADIF Deinterlacing, I am getting extremely bad experience when using Jump Forward or Jump Back or Fast Forward or Rewind (either remote/keyboard/or auto commercial skip):
-first get several seconds of video frozen on screen while audio continues to play
-then additional time of video frozen but no audio
-then see video quickly ff to "catch up" still no audio
-after video "catches back up", the audio starts to play
-video playback normal until using Jump/FF

Larger the jump => larger amount of time to recover. And simply pressing Stop will trigger MC to just hang for several seconds before going back to theater view, or crash MC completely (closes no error).

Unfortunately in it's current version (30.0.24), I cannot use JRVR YADIF DI.
I am playing TS files (1080 & 720 mpeg2 or mpeg4 recorded tv) 4k@59Hz hdmi & bitstream audio.
GPU is nvidia 1660ti 6GB
CPU and GPU numbers unchanged with DI enabled vs disabled.
JRVR settings:
Image Scaling = spline 3 taps
Chroma Scaling = cubic shader 150
Doubling = disabled
Debanding = disabled
Dithering = ordered

[jmone]

Ugg, retested it .... and I'd left my test PC in Custom Mode .  Put it back to std Red October JRVR and I'm seeing the similar issues as tzr916 with seeking.

[Hendrik]

That should be fixed in the next build.

[jshiu]

Hi, is it possible to install mc30 to a microsoft surface tablet which uses window?