xy-VSFilter Project (High Performance VSFilter Compatible Subtitle Filters)

[KoD]

A quick question:

I've noticed that people said xysubfilter can be used with DXVA native. However, I don't seem to be able to use DXVA native, in ZoomPlayer, when using LAV video for decoding H264, xysubfilter, and madVR v0.86.10. As soon as I unregister xysubfilter, DXVA native works (there are no other subtitle filters registered on the system).

I have disabled "Smart play" in ZoomPlayer to be able to see subtitles. xy-vsfilter is unregistered, so only xysubfilter is registered. XYSubFilter is configured with the default settings.

Are there any specific requirements to be able to use DXVA native (in ZoomPlayer) ?

[Pulstar]

Originally Posted by clsid

Yes. However, the subtitles will be rendered at the resolution of the video (expanded to 16:9) while the ISR can render at the resolution of your screen. So subs may be a bit less sharp for low resolution videos.

@cyberbeing
Can an "expand to 16:10" option be added to xy-VSFilter? Many people have monitors with such an AR.

I'm reverting to xy-VSfilter because of this.

[turbojet]

ASS subs were the reason for style not working, would be nice to force a style on them in the future. Styles are working fine on other text subs.

Thinking way ahead would there be any consideration in resizing image based subs (sup, vobsub, etc.)? Bdsup2sub can do this but it seems like a lot more work than is necessary.

[Moragg]

If possible, a feature I'd like is to turn the horrible yellow vobsub into something better - like a simple white font.

[Soukyuu]

I'm not quite sure, but isn't vobsub basically bitmaps? I don't think you can restyle them.

[Moragg]

Okay, I read up a bit and it looks like each subtitle is rendered as its own bitmap - didn't think anyone would do that as it seems horribly inefficient.

Maybe incorporate some on-the-fly character recognition and replacement? Or - slightly (a lot?) easier to do - force vobsubs to be rendered in black and white. What I hate most of all is the yellow colour, get it white and I'd be much happier.

[turbojet]

They are images so no chance to change font but the colors are controlled by a palette thats configurable in *vsfilter. I'm not sure it works, nothing handy to test atm but vaguely remember mpc isr working fine to change colors.

Scaling options move image subtitles, the only thing that's missing that should be possible is resizing/zooming.

[onomatopellan]

Are there any known plans to make more xySubFilter subtitle consumers? As I can't use madVR because of my intel integrated graphics I'm using latest MPC-HC with LAVFilters and EVR-CP as video renderer and I'd like to use xySubFilter.

[filler56789]

Originally Posted by Moragg

If possible, a feature I'd like is to turn the horrible yellow vobsub into something better - like a simple white font.

Just for the record --- there ARE people who think quite the opposite, i.e., "simple white fonts" do not have sufficient visual contrast, and we prefer yellow subtitles to white subtitles. No, we don't like to manually edit the color palette in the .IDX files, but this is what we have to do now and then.

[cyberbeing]

Originally Posted by thewebchat

Change flush does absolutely nothing. I see the subtitle queue instantly drop to 1 as soon as the first subtitle event activates. I observe average 8% CPU usage windowed and 30% CPU (one core completely saturated) in fullscreen.

CPU: Intel Ivy Bridge M @ 2.3 GHz
GPU: Nvidia GT650M
OS: Windows 8

Hopefully we can find a way to optimize the effect in this script a bit more.

Originally Posted by thewebchat

Regardless of my hardware, the subtitle renderer should never cause the video renderer to drop frames. If the subtitle renderer can't meet its deadlines, it should drop *subtitles* instead of causing unwatchable video.

Open a bug report on madshi's bugtracker. madVR is in charge of requesting and displaying subtitle bitmaps, so you have that reversed. If madVR cannot meet it's deadlines it should either A) not be requesting new subtitle bitmaps or B) dropping existing subtitle requests in favor of not dropping video frames. Currently madVR is likely designed to drop video frames when subtitle request are late, so you'd need to request this be changed.

Originally Posted by thewebchat

Edit: I don't observe any performance problems when pre-scaling the subtitles to 2560x1440 using Aegisub. The example file http://www.sendspace.com/file/iybf50 plays without any issues while the original file chokes Xy.

You've not scaled \be in that sample though. If you replace every \be1 with \be3 or \be4 does it still play fine?

Originally Posted by KoD

I've noticed that people said xysubfilter can be used with DXVA native. However, I don't seem to be able to use DXVA native, in ZoomPlayer, when using LAV video for decoding H264, xysubfilter, and madVR v0.86.10.

This problem cannot be caused by XySubFilter, since we never connect to a video pin or do anything else which would interfere with DXVA. At least that we are aware of. If DXVA is not working, it must be a madVR bug or ZoomPlayer bug.

Originally Posted by turbojet

ASS subs were the reason for style not working, would be nice to force a style on them in the future. Styles are working fine on other text subs.

You were using xy-VSFilter (VSFilter.dll) then? The reason this was disabled in xy-VSFilter is because the existing VSFilter override behavior for how this was triggered and applied to ASS subtitles was rather annoying.

Have you tested our new style override dialog for ASS subtitles in XySubFilter?

Originally Posted by turbojet

Thinking way ahead would there be any consideration in resizing image based subs (sup, vobsub, etc.)? Bdsup2sub can do this but it seems like a lot more work than is necessary.

In xy-VSFilter (VSFilter.dll) no, there are no plans to add resize functionality for image-based subtitles.

With XySubFilter, madVR is in charge of resizing all image based subtitles.

Originally Posted by Moragg

Okay, I read up a bit and it looks like each subtitle is rendered as its own bitmap - didn't think anyone would do that as it seems horribly inefficient.

Exactly the opposite, multiple smaller bitmaps is often more efficient than combining everything into one massive bitmap.

Originally Posted by Moragg

Maybe incorporate some on-the-fly character recognition and replacement?

What do you mean by this?

Originally Posted by Moragg

Or - slightly (a lot?) easier to do - force vobsubs to be rendered in black and white. What I hate most of all is the yellow colour, get it white and I'd be much happier.

Originally Posted by Moragg

If possible, a feature I'd like is to turn the horrible yellow vobsub into something better - like a simple white font.

Possibility for the future.

In the meantime, you could always force a custom style for VOBSUBs by enabling "custom colors" in the IDX file and editing the style to your preferences. This is something VSFilter has always supported, but rarely used for some reason.

Originally Posted by onomatopellan

Are there any known plans to make more xySubFilter subtitle consumers?

No known plans at the moment, but MPC-HC at least seems to be considering the possibility.

[cyberbeing]

MPC-HC has merged a change into the latest nightly to allow both xy-VSFilter (VSFilter.dll) & XySubFIlter to co-exist a bit better. This change will block VSFilter.dll from being used with madVR when XySubFilter is installed, while still allowing VSFilter to remain installed and be used with other video renderers. MPC-HC nightly builds can be found here

[sneaker_ger]

I kinda liked the older revisions better: just block xy-vsfilter under external filters if you want to use XySubFilter, or set to "prefer" if you do in fact want to use it. Now, if you want to quickly switch to xy-vsfilter for a test you have to actually un-register XySubFilter. Well, it's a good enough quick fix for beginners...

Originally Posted by Moragg

Okay, I read up a bit and it looks like each subtitle is rendered as its own bitmap - didn't think anyone would do that as it seems horribly inefficient.

Originally Posted by cyberbeing

Exactly the opposite, multiple smaller bitmaps is often more efficient than combining everything into one massive bitmap.

A typical bitmap based subtitle for a cinema movie is maybe a few Megabytes in size. This is still an order of magnitude bigger than text based ones (unless you stuff your mkv with fonts!), but compared to the video and audio data it's almost negligible. Or maybe he means "one bitmap for every video picture" instead of "for every line"? The former would indeed be more inefficient, but DVD and Blu-Ray use the latter, i.e. one subtitle line that spans over 10 seconds uses only one bitmap. At least Blu-Rays even support simple effects like fading for PGS subs without creating a new picture for every frame of the effect. Also, some RLE compression is used (and zlib in mkv, though optional).

Originally Posted by cyberbeing

What do you mean by this?

He means real-time OCR, i.e. internally converting to text based subs and then rendering with whatever style (font, color,...) you want. Everyone who has some experience with converting bitmap based subs to text based ones will know that this would be a very ambitious feature - even the programs that are specifically designed for the job and have had years of development still require lots of user-interaction regularly, so it's not really suited for full automation/real-time (at least with existing free libraries).

[cyberbeing]

Originally Posted by sneaker_ger

I kinda liked the older revisions better: just block xy-vsfilter under external filters if you want to use XySubFilter, or set to "prefer" if you do in fact want to use it. Now, if you want to quickly switch to xy-vsfilter for a test you have to actually un-register XySubFilter. Well, it's a good enough quick fix for beginners...

Well this is one of those things where if someone installs XySubFilter, there is an expectation they actually want it to be used with madVR. If you just want to see what something would look like with VSFilter.dll style rendering, you could always enable the "Render to Original Video Size" option in XySubFilter.

Maybe you could request that MPC-HC add a registry key to toggle this behavior, like the hidden key they use to toggle the ISR blocking VSFilter behavior.

Originally Posted by sneaker_ger

one subtitle line that spans over 10 seconds still uses only one picture.

The subtitle interface actually works in this way already. Each bitmap is given a unique id as well as a start/stop time.

We also have an advanced multi-level caching mechanism in xy-VSFilter/XySubFilter which makes rendering identical portions of text subtitle lines use nearly zero CPU time. This caching is what makes Libass so much slower than xy-VSFilter at times. Libass is forced to re-render identical computational heavy operations every frame, while we render once and then pull it from our cache for future frames.

Originally Posted by sneaker_ger

He means real-time OCR, i.e. internally converting to text based subs and then rendering with whatever style (font, color,...) you want. Everyone who has some experience with converting bitmap based subs to text based ones will know that this would be a very ambitious feature - even the programs that are specifically designed for the job and have had years of development still require lots of user-interaction regularly, so it's not really suited for full automation/real-time (at least with existing free libraries).

...yeah, probably not going to happen. VSFilter traditionally has always converted VOBSUB bitmaps to vector though.

From what I can tell, all image based subtitles formats (VOBSUB/PGS/DVB) have a color palette, but the hard part would likely be knowing which color of the palette to modify. Overall since our focus has mainly been on text subtitles, any features such as this would best be requested of MPC-HC or MPC-BE who actually work on improving the code for image based formats.

[sneaker_ger]

Originally Posted by cyberbeing

Well this is one of those things where if someone installs XySubFilter, there is an expectation they actually want it to be used with madVR. If you just want to see what something would look like with VSFilter.dll style rendering, you could always enable the "Render to Original Video Size" option in XySubFilter.

Maybe you could request that MPC-HC add a registry key to toggle this behavior, like the hidden key they use to toggle the ISR blocking VSFilter behavior.

Yes, that would be nice maybe. But I know this is still being discussed and I'm currently sticking to an older MPC-HC revision anyways because the current ones override LAV Filter's stream selection logic.

[cyberbeing]

How to take madVR screenshots with MPC-HC which include XySubFilter subtitles

[cyberbeing]

Yeah vivan, as you discovered yourself in your edit, both .\sub unable to be created and the default path entries unable to be deleted are behaviors inherited from VSFilter which have never been fixed or modified.

Originally Posted by vivan

Also I would like to remove "c:\subtitles" path, but Remove button is disabled.

No idea why this is, but this was what Gabest set as the default behavior when he originally created VSFilter and nobody ever thought to change it. Eventually, this legacy search path behavior will likely be scraped and replaced with something more feature rich. Until then, it is how it is.

[Moragg]

Originally Posted by filler56789

Just for the record --- there ARE people who think quite the opposite, i.e., "simple white fonts" do not have sufficient visual contrast, and we prefer yellow subtitles to white subtitles. No, we don't like to manually edit the color palette in the .IDX files, but this is what we have to do now and then.

I normally watch anime, and yellow just doesn't work - since the backgrounds are often bright and have a similar texture it clashes badly (in my eyes) and makes it harder to read.

The argument is a lot stronger for live-action movies, but even there I find them too intrusive, and white wouldn't upset my viewing as much. But here I would also say that yellow could be a better option in many circumstances.

Originally Posted by cyberbeing

Exactly the opposite, multiple smaller bitmaps is often more efficient than combining everything into one massive bitmap.

But much less efficient than a font and text and video.

Originally Posted by cyberbeing

In the meantime, you could always force a custom style for VOBSUBs by enabling "custom colors" in the IDX file and editing the style to your preferences. This is something VSFilter has always supported, but rarely used for some reason.

Didn't know about this, will definitely check it out, thanks.

Originally Posted by cyberbeing

...yeah, probably not going to happen.

Shame, didn't realise it was that difficult to do.

[cyberbeing]

Originally Posted by Moragg

But much less efficient than a font and text and video.

Depends on the font, some complex font outlines are extremely computationally heavy to render.

[Telion]

Can you please implement a feature to save & reopen last selected properties tab? It will be especially handy considering the current workaround for taking screenshots with subtitles. Timing corrections will also benefit from this.

[cyberbeing]

We'll look into it, I created Issue #158. In the future, feel free to make such feature requests directly on our bug tracker.