You are not logged in. Please login or register.
SmoothVideo Project → Posts by mark007
Thanks a lot for that. Are those functions available within SVP (ie some BFI option) or would I need to create my own vapoursynth script?
I am using an LG C1 at 120hz and have recently being enlightened by various YouTube channels showing how to turn BFI on in the display itself. However it is not configurable in terms of how long to show the black frame. So it works well if showing 120fps at 120hz as the display refreshes at 240hz, half being black, half being the image. No flicker is noticeable here.
If I use BFI at eg 60hz or lower like 24hz the flicker is too annoying as the black frame is on screen for so long. I wanted to experiment with a software based mechanism for showing black frames for a configurable amount of time.
I would love to try SVP with 24fps content on a 120hz screen and try for example 3 out of every 5 frames showing the image, 2 out of every 5 showing black. Maybe even 4 out of 5 showing the image, 1 being black. To see which increases motion resolution with least noticible flicker.
Can this be done with SVP. I'm using SVP with MPC, madVR and vapoursynth and want to try without interpolation but simply BFI as I can see even a 60fps game with BFI looks as smooth as a 120hz FPS game.
Issue 1 - Freeze after a few frames at start off playback
Yes, it looks 0.8.1 is working better. I have not seen the issue yet 
Edit: After a few reattempts, its now occurring also with 0.8.1. The frame its frozen on is I would estimate 8-10 frames into the scene. You can see the audio continues from the seek bar still moving. cpu usage has dropped to almost 0 as theres nothing happening from a video point of view. I have now reproduced without SVP even switched on, just by having avisynthfilter in the graph is enough to trigger it. I'll log it on the avisynthfilter github page.
Issue 2 - Seeking causes old frames to reappear / flicker during playback
The second issue thats currently not allowing me to use SVP is the issue mentioned above in a previous thread, where it looks like frames from a few seconds ago are still cached / queued or something. So the old images flicker in and out, even with this latest avisynthfilter 0.8.1. Would that be an issue in the filter or in SVP or in both? I'm mainly seeing this in HDR content when in fullscreen. I can't catch this in a screenshot, only a picture taken from my phone, or by frame advancing in mpc, by doing Ctrl-Right. You can see normal frames appear, then soon followed by some frames from before the seek even happened. Like as if they needed to be flushed from some buffer but were not, and so turn up in the playback.
Hi, another issue, I assume also somehow related to the frame queue/cache. When madVR is set (as I have it set) to "delay playback start until render queue is full", and "delay playback start after seeking, too", sometimes (perhaps 50% of the time, but I can't find a pattern), when starting some videos, the image plays a few frames and then freezes. I have gpu and cpu queue set at 16 now. The audio continues in the background fine.
Manually seeking in the file, restores video playback.
Changing display modes (by double clicking madVRs image and letting it change my desktop from default of 60 to 48hz) seems to completely stop the audio.
This clip is an interlaced clip that I'm using LAV Video Yadif to de-interlace to double its original frame rate. (its a live performance of Lady Gaga Born This Way titled "Born This Way - 2011(HDTV-1080).mpg")
Is it something to do with SVP initially turning on, but then a few ms into the clip, it is switching to the 'Ignore HFR sources' profile as the clip is high frame rate (after being deinterlaced). Is this switch somehow causing the issue.
Heres the event log.
18:38:29.549 [i]: AVSF: found new player instance
18:38:29.592 [i]: AVSF: filters in use: LAV Splitter Source -> LAV Video Decoder -> * -> madVR Renderer
18:38:29.593 [i]: AVSF: new video in mpc-be64.exe (64-bit) [MPC-BE x64 1.5.5.5274] on screen 0
18:38:29.662 [i]: Media: video 1920x1080 [PAR 1.000] at 29.970 fps
18:38:29.662 [i]: Media: codec type is MPEG Video, YUV/4:2:0/8 bits/BT.709 BT.709
18:38:29.663 [W]: Media: frame rate info can be incorrect - 59.940 fps via video player vs. 29.970 fps via MediaInfo
18:38:29.663 [i]: Media: scan type is interlaced, assuming 59.940 fps instead of 29.970 fps
18:38:29.664 [i]: Playback: starting up...
18:38:29.671 [i]: Playback [ff91ef3c]: Frame server (64-bit) 3.5.0.0, AviSynth+ 3.5 (r3106, 3.5, x86_64), C:\WINDOWS\SYSTEM32\avisynth.dll
18:38:29.673 [i]: Playback [ff91ef3c]: resulting video frame 1920x1080
18:38:29.673 [i]: Playback [ff91ef3c]: 2 acceptible profiles, best is 'Ignore HFR sources' [592]
18:38:29.675 [i]: Playback [ff91ef3c]: enabled while video is in some unknown state
18:38:29.675 [i]: Profile: using automatic profile #2 /film
18:38:29.681 [i]: Playback [ff91ef3c]: playing at 59.94 [59.94 *1/1] Its a blu-ray that I ripped onto my hard drive, its an m2ts/index.bdmv etc. Playing via the index or m2ts file both show the fact that its too short. Perhaps its some issue with the MediaInfo of the file, or how does SVP determine the length. I can check those details locally and share.
I just purchased the pro version and am finding whats the best settings for 4k HDR at the moment. I'm using the avisynth filter. I'm finding when seeking sometimes, the image seems to "remember" or cache some frames and it seems forever, it randomly repeats those images over and over, in a flickering manner. I would assume the default SVP Pro settings should not behave like this? Is it related to the avisynth filter? I remember in the past when using the ffdshow method to invoke SVP, I used to have to tweak the buffers to prevent this from happening. Is something similar required here in the avisynth filter?
This is when in madVR (the latest beta, set to output HDR to my TV, LG C9), and in fullscreen windowed mode. In smaller windowed mode it doesn't seem to occur. I have all of madVRs buffers set to 16 frames, both cpu and gpu.
I just purchased the pro version and have started testing playback on my machine, with mpc-hc, lav video, the avisynth directshow filter.
I have noticed some of my files also require me to set the minimum duration to 0 even though they are full length Movies. I have been playing the bond movie, 'The World Is Not Enough'. It reports.
Playback: ignoring too short clip (3 secs)
Are there any logs I could send to see why this is required. Setting min duration to 0 is fine for me, just; I'm wondering that others may also be affected.
Hi,
I use my own custom variation of InterFrame scripts with SVP 3.1 currently and would love to move to SVP 4 if there are any quality improvements?
I tried using the two svpflow1.dll and svpflow2.dll files from SVP 4 Free but my existing script fails with the error below.
"Script error: SVSmoothFps does not have a named argument 'url'"
Is this possible with SVP 4 Free, or with SVP 4 paid or is there a different set of arguments to SVSmoothFps?
Cheers,
Mark
MAG79, is there any plans in the near future to move more load currently done on CPU over to the GPU? Is it a big development effort to achieve this?
I'm like many of the guys on doom9 with a shiny new GTX 980 for my HTPC but currently SVP hits CPU 100% bottleneck when increasing quality where as GPU is still only using very minimal %, and its fans don't even spin up 
But do you use SVP manager or custom script that uses SVP DLLs. I use custom script and find varying ffdshow buffers show different audio sync results across all of machines
I guess SVP manager sets one fixed buffer ahead value?
I have vsync forced on. No setting should ever cause audio to go out of sync but for me ffdshow settings are all that ever effect it so there's definitely an issue with it.
If you are still having the issue I'd suggest to try custom script and play with buffer ahead values. At least that was a good workaround for me and shows the root cause of the issues at least for my machines is ffdshow buffer ahead values.
No. On doom9 forums as well as this thread itself 
and many more threads in this forum reporting audio out of sync where resources arnt the issue.
How could very specific ffdshow buffer ahead cause audio desync when one value above or below doesn't... if its not some issue between ffdshow and avisynth...
I welcome you to also play with ffdshow buffer values to reproduce. I found 15 to be perfectly in sync for my current setup so I leave it as is but still the question remains why does it cause audio desync if its not tuned like I did. GPU usage on my system is extremely low. The fans don't even spin. CPU usage is around 40%
mark007
somehow running avisynth without ffdshow?
You can Generate AVS script by SVP menu - for video encoder. Then Close SVP. And then open this AVS file in your player.
Oh wow. But how can I use avisynth while playing a DVD or bluray through mpc for example. How can say lab decoder feed into avisynth?
Razelik. No I have changed hardware over the years amd thr issue remains.
Currently GTX 980 with core i7 @ 3.6ghz. There's definitely issues with ffdshow interacting with avisynth when it comes to thread counts and buffer ahead sizes so in hope to elimate ffdshow from filter chain but not sure how to still use it without ffdshow without comprises
I have given up on hope of a fix.
Nobody must read my posts about audio going out of sync. For me it never ever has anything to do with reclock or renderers, but always avisynth / ffdshow setup. Try changing ffdshow buffer ahead settings. See here http://www.svp-team.com/forum/viewtopic.php?id=1835
Nobody seems to know whether its going to be possible to fix these issues as ffdshow is pretty much end of life at this point. MAG79 any idea if these can be fixed by somehow running avisynth without ffdshow?
I get audio desync also and its usually caused by ffdshow buffer values. With my core i7 920 and avisynth from SET I use ffdshoe buffer values of 0 and 15 and also use cores value in SVP of 15. Try playing with ffdshow buffer ahead values.
FYI SEt posted his new MT version http://forum.doom9.org/showthread.php?t=148782
Thanks very much. Great
I seem to get no banding when disabling the new linear option. Is it caused by this out of interest.
Cheers.
Looks like clsid doesn't know the ffdshow aviaynth code. Can SVP ever work outside ffeshow or not need ffdshows buffers? With buffers off video is perfectly smooth for me but desync is really bad
Yeah I asked in the ffdshow thread, hopefully clsid might get back, hopefully its something they can spot and fix easily
Oh wow OK. Is this something I can bring to them or as a dev do you want to bring to them
Cheers that sounds great. Would you have any contacts that you feel might be interested to fix it.
Looks like only buffer effects the timestamps. Whats the next step in narrowing down the source of the issue
Do you think this is ffshow avisynth SVP or some combination of all. Its probably a missing +1 or -1 somewhere I wish I could help more but all I can do is report my findings.
OK I used ffdshows OSD feature to output details about timestamps, frame numbers, and asked it to write to file the details about the first 100 frames.
Using the same script to interpolate by eg 60/24, the output should be the same each time. If I keep the buffers / threads the same, the timestamps remain the same. If I change them the timestamps change.
To me this is a 100% indication that there is some logic problem somewhere. The timestamps shouldn't change at all, they should always be the exact same no matter how many threads or buffers are used. Otherwise its a bug.
Oh I know, and the players / renderers even bundled with SVP do a great job of keeping them in sync (mpc, madVR, LAV, reclock etc) once the timestamps are correct. But if they are fed incorrect timestamps they have no hope, and we need to add workarounds of adjusting audio delay etc which isn't the right approach.
Is there nothing that can be done without avisynth / ffdshow developers to look into this?
SmoothVideo Project → Posts by mark007
Powered by PunBB, supported by Informer Technologies, Inc.