[Ffmpeg-devel-irc] ffmpeg.log.20200116

burek burek at teamnet.rs
Fri Jan 17 03:05:01 EET 2020


[00:07:36 CET] <cehoyos> Attached images currently only work for flac and mov, this is not a trivial feature to add.
[00:09:18 CET] <cehoyos> It also works for mp3 using different apis
[00:13:52 CET] <fatturtle> Thats too bad. Oh well.
[00:18:56 CET] <furq> fatturtle: opusenc supports flac input and cover art
[05:57:01 CET] <void09> anyone here knowledgeable about bluray's m2ts format?
[05:58:32 CET] <void09> I was wondering how hard it would be to reconstruct an original m2ts file from an mkv remux of it. what would you need to backup from it, in order to reconstruct it (except the actual audio/video/sub stream) ?
[05:58:46 CET] <void09> bit perfect reconstruction that is
[09:04:18 CET] <kurosu> michaelni: if (get_bits_left(gb) < width) return AVERROR_INVALIDDATA; <- there are probably several codecs that would benefit from moving such checks to a higher-level loop, if possible (thanks captain obvious)
[09:04:27 CET] <kurosu> but speed-vs-security trade-off
[09:04:48 CET] <kurosu> ouch wrong channel XD
[12:12:08 CET] <JulietDeltaGolf> hi !
[12:35:49 CET] <JulietDeltaGolf> I'm converting a 2K DPX sequence to FFv1. Is there some option I'm missing on the cmdline to speed things up ?
[12:36:13 CET] <JulietDeltaGolf> ffmpeg.exe -framerate 24 -i "frame_%06d.dpx" -c:v ffv1 -coder 1 -context 1 -g 1 -level 3 -threads 64 -slices 64 -slicecrc 1 -f matroska NUL
[12:36:47 CET] <JulietDeltaGolf> looks like a part of the encode is single threaded as I do see one core close to a 100%
[12:37:18 CET] <JulietDeltaGolf> also why does ffmpeg do not recommend going above 16 threads for ffv1 ?
[12:37:49 CET] <durandal_1707> dpx decoder is not supporting multiple threads
[12:38:15 CET] <durandal_1707> support could be added ...
[12:38:36 CET] <JulietDeltaGolf> oh ...
[12:39:15 CET] <JulietDeltaGolf> but isn't it just a matter of opening the file and putting the content in memory isn't it ?
[12:39:25 CET] <JulietDeltaGolf> a single core should be able to do more than 12 fps ...
[12:40:26 CET] <durandal_1707> JulietDeltaGolf: try with only dpx decoding and compare
[12:41:00 CET] <JulietDeltaGolf> ffmpeg.exe -framerate 24 -i "frame_%06d.dpx" -c:v r210 -f matroska NUL
[12:41:03 CET] <JulietDeltaGolf> would this do the trick ?
[12:41:19 CET] <JEEB> -f null -
[12:41:37 CET] <durandal_1707> just do: ffmpeg.exe -framerate 24 -i "frame_%06d.dpx" -f null -
[12:41:42 CET] <JEEB> yea that
[12:41:58 CET] <JEEB> that way it will utilize whatever data comes out of the decoder
[12:48:30 CET] <JulietDeltaGolf> 24 fps
[12:48:48 CET] <JulietDeltaGolf> average 70% one one core
[12:51:59 CET] <durandal_1707> afaik if decoder is single threaded it will use single thread even for encoding
[12:58:15 CET] <DHE> no you can have a single threaded decoder and multithreaded encoder. ffmpeg doesn't care...
[13:00:25 CET] <JulietDeltaGolf> so if my single threaded dpx decoder can push 24 fps of content into the pipeline
[13:01:11 CET] <JulietDeltaGolf> is there something I can do to speed up ffv1 encoding then ?
[13:05:46 CET] <durandal_1707> DHE: but performance will suffer greatly as experiment here shows
[13:06:24 CET] <durandal_1707> JulietDeltaGolf: pay some developer to speed up dpx decoder
[13:07:23 CET] <JulietDeltaGolf> yeah apparently I'll have to speed up matroska wrapper as well
[13:07:46 CET] <JulietDeltaGolf> dropping the wrapping got me from 11 fps to 16 fps
[13:47:09 CET] <familiyaF> I am writing an application using libav with CUDA transcoding. is there a way to determine if the input pixel format is supported by cuvid or not
[13:49:00 CET] <JEEB> the encoders have a pix_fmts listing
[13:49:07 CET] <JEEB> but the problem is that it probably depends on the hardware etc you have
[13:49:33 CET] <JEEB> so the encoder wrapper might say it supports something, but with old enough drivers or hardware it might not
[13:49:36 CET] <JEEB> :P
[13:49:43 CET] <JEEB> this will get caught when you try to open the encoder most likely or so
[13:50:34 CET] <familiyaF> ok
[13:50:36 CET] <Azlux> Do there are any memory leak issue on ffmpeg ? User of my bot experience it since they have upgrade to last debian verison : 4.1.4-1~deb10u1.
[13:50:36 CET] <Azlux> This bot use this command : https://github.com/azlux/botamusique/blob/master/mumbleBot.py#L668
[14:44:33 CET] <pagios>  -strict -2 -crf 18 <-- where can i read aboutthose params
[14:48:49 CET] <DHE> when in doubt https://ffmpeg.org/ffmpeg-all.html has literally everything, but it's a huge document
[14:53:54 CET] <cehoyos> Azlux: Many bugs exist, please test current FFmpeg git head and provide valgrind output for memleaks
[14:54:52 CET] <cehoyos> pagios: I don't think you can read about "strict -2" (or actually I don't think the answer would help you) but if you provide the command line you tested together with the complete, uncut console output, we may be able to help you
[14:55:20 CET] <pagios> waht is strict -2 :)
[14:55:22 CET] <pagios> and -crf
[14:56:32 CET] <DHE> -strict -2 unlocks options, parameters and codecs that either break the normal rules/standards of video encoding or lets you use experimental/in-development codecs and features
[14:57:19 CET] <JEEB> -strict experimental is the non-number version of that
[15:02:00 CET] <pagios> any recommended books to understand basics for video/audio? would make more sense then if i want to learn to tune
[15:09:47 CET] <pagios> JEEB,  DHE? i would like to learn the basics of video and all this stuff , really back to basics
[15:29:53 CET] <ocx32> hi all! i am trying to optimize hls latency i am using  -preset:v ultrafast -hls_time 1 -hls_list_size 10 <--- can i still optimize it more even if create more disk io? want to lower latency thanks!
[15:32:47 CET] <DHE> HLS just isn't intended to be a low latency format. even that will give you ~2 seconds of delivery latency at very best.
[15:33:18 CET] <ocx32> DHE yea i am aiming for 2 3 sec
[15:33:29 CET] <ocx32> i am getting around 20sec with that config
[15:33:36 CET] <ocx32> what can i optimize
[15:34:09 CET] <DHE> what's your full ffmpeg commandline? (censor passwords or anything sensitive with "XXXX" or whatever, but omit nothing)
[15:35:06 CET] <ocx32> DHE: -i rtmp:/.... -vf scale=1280:720 -vb 2500k -g 90 -framerate 30 -preset:v ultrafast -hls_time 1 -hls_list_size 10 -f hls /..../x.m3u8
[15:35:40 CET] <DHE> okay, -g 90 means a keyframe is every 90 frames. at 30fps that's 3 seconds between keyframes. and that is effectively overriding your -hls_time 1 parameter
[15:36:09 CET] <ocx32> oh!
[15:37:03 CET] <bencoh> /41/59
[15:37:07 CET] <bencoh> (woops)
[15:38:32 CET] <ocx32> DHE anything else i can optimize?
[15:39:36 CET] <DHE> if it really really matters you can try -tune:v zerolatency  but there will be an image quality hit
[15:39:49 CET] <DHE> but with ultrafast mode it's not likely to have as much of an effect
[15:40:33 CET] <ocx32> ultrafast is ok
[15:41:10 CET] <ocx32> i mean the quality i am getting now
[15:47:03 CET] <ocx32> i am still getting 10sec , a bit better
[15:49:44 CET] <ocx32> DHE: actually the -g 30 when removed didnt make any difference, i jsut noticed
[15:50:38 CET] <bencoh> -g 30 or -g 90 ?
[15:50:52 CET] <DHE> and what do you mean "removed" ?
[15:56:33 CET] <ocx32> DHE i mean i removed from the command the -g 90
[16:01:16 CET] <DHE> ocx32: put it back and set it to 30
[16:01:36 CET] <ocx32> DHE i just did this 5 minutes ago :)
[16:01:47 CET] <ocx32> still same results
[16:09:53 CET] <ocx32> any idea?
[16:18:29 CET] <ocx32> DHE?
[16:45:24 CET] <DHE> sorry am busy
[16:48:19 CET] <ocx32> sure
[17:59:43 CET] <lofo> i'm trying to debug my command ffmpeg -f avfoundation -i "0" -f rtp_mpegts -c:v libx264 rtp://192.168.1.47:1234
[18:00:06 CET] <lofo> i get a ton of `Non-monotonous DTS in output stream 0:0; previous: 552, current: 50; changing to 553. This may result in incorrect timestamps in the output file.` on console output
[18:59:24 CET] <CpAj0> hello, i need to create a new video file file given one or more segment of duration, is possible to do with ffmpeg?
[19:42:39 CET] <antto> is there a way to rotate by 180° (or flip vertically and horizontally) a video quickly with minimum loss?
[19:43:46 CET] <BtbN> Some containers have flags instructing the player to rotate on playback
[19:43:52 CET] <BtbN> But other than that, not really
[19:43:52 CET] <antto> i've captured it like that due to camera/tripod setup, but now if i put flip X/Y or transform effects in kdenlive to fix it, the preview becomes very slow, i'd like to "preprocess" it maybe
[19:44:20 CET] <BtbN> It will be a fully re-encode, which comes with quality loss
[19:44:23 CET] <BtbN> -y
[19:44:28 CET] <antto> right
[19:45:23 CET] <antto> "ffmpeg -i in.mp4 -filter:v "transpose=1,transpose=1" out.mp4" <- like this?
[19:46:52 CET] <SoMuchForSubtlet> Is there a simple way to save the highest resolution video track and all audio tracks (or the audio track in a specific language) from a master m3u playist like this? https://pastebin.com/raw/bZPBB3sV I know I can use ffprobe to find the desired tracks, but I want to know if there is a simpler and quicker way. Thank you
[19:47:38 CET] <durandal_1707> antto: flip use hflip/vflip filters
[19:48:08 CET] <Hello71> I don't see how that would be different from transpose,transpose
[19:48:19 CET] <antto> whichever runs faster
[19:48:48 CET] <antto> i know a rotation algorithm is gonna be brutal, while a mere flip is gonna be much easier
[19:49:04 CET] <Hello71> well you definitely shouldn't use rotate
[19:49:40 CET] <antto> yeah, but it seems like in kdenlive transform and flip result in roughly equal "slowness"
[19:49:58 CET] <antto> so you never know ;P~
[19:51:28 CET] <antto> ffmpeg -i in.mp4 -vf "hflip,vflip" out.mp4 ?
[19:54:11 CET] <antto> the video is also not 1:1 pixel aspect
[19:55:41 CET] <BtbN> What does the -ar cli option do exactly? Automatically insert an aresample filter?
[20:04:01 CET] <BtbN> From how I understand the loudnorm documentation, it by default, in linear mode, should not resample to 192k?
[20:04:11 CET] <BtbN> The code is very unclear about the whole resampling
[20:16:32 CET] <matto312> Hello! I've been stuck on an issue for a long time and I just can't get past it ... I have live RTMP input and multi-variant HLS out ... sometimes when I start FFMPEG the screen gets flooded with "DTS 4294991118, next:2147507469961 st:1 invalid dropping" & "PTS 4295057956, next:2147574306250 invalid dropping st:1" messages ... I'm pretty desperate, just any direction on what to look at it would be so much appreciated.
[20:16:52 CET] <matto312> like any thoughts at all on where to investigate
[20:24:02 CET] <matto312> Just like any random thought that comes to your head that could help me out
[20:24:15 CET] <matto312> i'm desperate I will send money
[00:00:00 CET] --- Fri Jan 17 2020


More information about the Ffmpeg-devel-irc mailing list