[Ffmpeg-devel-irc] ffmpeg.log.20180917

[burek]

[01:15:16 CEST] <johnjay> 29.97...
[01:15:31 CEST] Action: johnjay wonders what standards documents like PAL or NTSC would be useful for reading ffmpeg
[01:59:45 CEST] <frendShot> for animation, is there anything specific I should look at when going from 30fps to 60fps using interpolation?
[04:21:55 CEST] <DHE> johnjay: probably be pretty horrific. the 29.97fps thing is a relic of colour TV to allow for backwards compatibility with B&W
[04:22:22 CEST] <DHE> in the digital world I would suggest avoiding NTSC and PAL unless you have to explicitly deal with them. and analog is being phased out in most of the world
[04:22:23 CEST] <johnjay> interesting
[05:49:59 CEST] <nicolas17> hi
[05:50:07 CEST] <nicolas17> the 'select' video filter supports scene change detection, doing stuff like select='gt(scene\,0.4)'
[05:50:28 CEST] <nicolas17> but can I *get* the scene change probability for *every* frame, rather than just selecting frames based on whether that probability exceeds a fixed threshold?
[05:50:29 CEST] <frendShot> for animation, is there anything specific I should look at when going from 30fps to 60fps using interpolation?
[05:50:47 CEST] <frendShot> like, to optimize it
[05:51:21 CEST] <frendShot> I noticed that with very busy and dynamic frame transitions, the interpolation looks all muddy and chaotic and loses detail
[06:03:40 CEST] <nicolas17> wow
[06:03:54 CEST] <nicolas17> the select filter does its own scd
[06:04:15 CEST] <nicolas17> I wonder how many scene change detection algorithms are in ffmpeg :/
[06:21:05 CEST] <nicolas17> yeah so every filter that does scene detection has its own code to do it
[06:21:13 CEST] <nicolas17> and they all just do sum of absolute differences
[06:26:22 CEST] <nicolas17> vf_decimate doesn't even use the optimized SAD version from pixelutils lol
[06:37:32 CEST] <nicolas17> yeah, meh @ getting all scd probabilities then
[06:37:54 CEST] <nicolas17> if it's just sum of absolute differences I can write that myself... I expected something fancier
[06:38:20 CEST] <nicolas17> like... frame differences after having done motion compensation
[08:31:15 CEST] <winny> Hi, I have this desktop recording script that seems to drop a lot of frames - https://ptpb.pw/1TSb.webm - looks completely smooth locally. Any tips on how I can record all frames? Here is my script: https://ptpb.pw/iAeE/sh
[08:34:43 CEST] <winny> oh sorry, please note the above has flashing colors
[08:35:02 CEST] <furq> it's probably dropping frames because libvpx-vp9 is slow as hell
[08:35:08 CEST] <furq> especially without row-mt
[08:35:35 CEST] <furq> also about half of that script isn't doing anything
[08:35:36 CEST] <winny> I could transcode later on
[08:35:54 CEST] <winny> indeed, it's a wip :s
[08:36:29 CEST] <ritsuka> yes, it would be better to use a lighter format that has a faster encoder available
[08:36:45 CEST] <winny> what encoder would you recommend for recording (and i'll do transcoding later on as suggested)
[08:37:14 CEST] <furq> try adding -row-mt 1
[08:37:24 CEST] <furq> but if that still drops frames then use ffv1
[08:37:49 CEST] <winny> ffv1 looks perfect for recoding source
[08:38:03 CEST] <furq> yeah ffv1 is a good intermediate codec
[08:38:12 CEST] <furq> the issue with libvpx is that it doesn't multithread well
[08:38:24 CEST] <winny> is there a recommended file extension for ffv1? does it matter?
[08:38:29 CEST] <furq> row-mt should hopefullt fix that
[08:38:35 CEST] <winny> i'll check out row-mt too
[08:38:40 CEST] <furq> mkv is fine for ffv1
[08:38:54 CEST] <winny> right, i guess I meant container :)
[08:40:13 CEST] <winny> ah maybe I need a newer ffmpeg, added -row-mt 1 after -threads 8 and get unrecognized option
[08:42:06 CEST] <furq> yeah it's fairly new in libvpx
[08:47:44 CEST] <winny> yeah, that helped a lot (using ffv1), but i think i could tweak it a little bit more: https://ptpb.pw/9YLP.webm https://ptpb.pw/RZzX/sh
[08:48:02 CEST] <winny> a lot of Past duration too large
[13:41:22 CEST] <mozzarella> guys help
[13:45:08 CEST] <durandal_1707> mozzarella: what?
[14:01:00 CEST] <DHE> did he died?
[14:02:20 CEST] <mozzarella> durandal_1707: I'm reencoding a video and it's extremely slow
[14:03:32 CEST] <BtbN> Buy more and faster cores
[14:03:41 CEST] <durandal_1707> mozzarella: what codec?
[14:03:49 CEST] <DHE> troubleshoot properly
[14:06:47 CEST] <mozzarella> durandal_1707: x264
[14:12:52 CEST] <DHE> are you seeing CPU pegging at 100% (or at least 96%)? probably, but it's worth checking to confirm where the
[14:14:03 CEST] <durandal_1707> mozzarella: which preset?
[14:14:35 CEST] <mozzarella> DHE: I'm limiting it to 25% using cpulimit, otherwise it's turning my room into an inferno, but it's slow relative to the cpu usage
[14:14:57 CEST] <Mavrik> Well.
[14:14:58 CEST] <BtbN> Can you talk numbers instead?
[14:19:58 CEST] <illuminated> ok... say you have a file with 2 audio streams.. 0 and 1.  0 is set as default.  you remap 1 to 0 and you want to set the new 0 as default... if you do a -disposition:a:0 0 and then do -disposition:a:0 default it doesn't unset the original default flag.. so the outputted file has 2 'default' tracks.
[14:20:23 CEST] <illuminated> am I doing something wrong?
[14:20:30 CEST] <illuminated> this is the behavior i'm experiencing
[14:23:26 CEST] <illuminated> the documentation on disposition would lead you to believe that to set a default disposition you use the stream numbers from the input file, but i've noticed the opposite in practice.. the default stream is dictated by the generated files stream numbers.
[14:24:32 CEST] <DHE> illuminated: it depends on the file format to a certain extent. some don't really have the ability to mark streams as "default"
[14:24:54 CEST] <DHE> one issue with ffmpeg is that it has to support such a wide variety of formats with a variety of features that these things do happen
[14:25:18 CEST] <illuminated> well the container is mkv and it has 2 audio streams and 2 subtitle streams
[14:28:14 CEST] <illuminated> if I have to I can check for this condition in the script and generate a temporary file and then run ffmpeg against it and set the right default streams... that's the only way I know to work around this behavior.
[15:33:17 CEST] <frendShot> for animation, is there anything specific I should look at when going from 30fps to 60fps using interpolation?
[15:33:20 CEST] <frendShot> like, to optimize it
[15:33:23 CEST] <frendShot> I noticed that with very busy and dynamic frame transitions, the interpolation looks all muddy and chaotic and loses detail
[17:11:00 CEST] <kepstin> frendShot: for any video, but animation especially, just avoid interpolating to higher frame rates completely?
[17:12:00 CEST] <kepstin> 30fps animation is already an unusually high rate for animation and should be pretty smooth.
[17:12:25 CEST] <frendShot> yeah, thing is this studio needs all videos uploaded at 60fps
[17:13:10 CEST] <frendShot> I tried to explain to them that our animators only do 24 and 30, but the guy is saying that they don't accept anything that isn't 60fps and that animation studios should "get with the times"
[17:13:16 CEST] <kepstin> if it's just a technical requirement, then duplicate frames (use the fps filter)
[17:13:39 CEST] <frendShot> the smoothness characteristic of higher framerates won't be there then, right?
[17:13:53 CEST] <kepstin> yeah, it'll look exactly the same as the original 30fps video
[17:13:54 CEST] <frendShot> because this stupid ad agency insisted on that
[17:13:59 CEST] <frendShot> ugh, yeah, that won't do
[17:14:02 CEST] <kepstin> eww.
[17:14:18 CEST] <frendShot> my boss is an idiot for signing a contract with them
[17:14:25 CEST] <frendShot> I don't know how the hell to deal with this
[17:14:51 CEST] <Mavrik> Tell the animators to do 60fps? :P
[17:14:54 CEST] <kepstin> if they insist on 60fps animation, then you might just want to go back to the animation studio and ask for more inbetween frames, yeah
[17:15:07 CEST] <kepstin> obviously this will increase the costs, but, well :/
[17:15:08 CEST] <Mavrik> Isn't that what you promised and sold to the customer? :)
[17:15:14 CEST] <frendShot> you want me to tell my artists and my animators to come up with more animation cells?
[17:15:18 CEST] <frendShot> and do this all over again?
[17:15:23 CEST] <frendShot> you want them to quit?
[17:15:34 CEST] <frendShot> they are not going to do any of that.
[17:15:56 CEST] <frendShot> Mavrik, that deal was done behind my back between the producer and this ad agency
[17:16:07 CEST] <frendShot> and my boss, the producer, was idiotic enough not to tell me anything about specifics
[17:16:12 CEST] <kepstin> well, motion interpolation isn't gonna help you if you want good looking video.
[17:16:17 CEST] <frendShot> when I go and send it to the ad agency, I get the "60fps pls" crap
[17:16:24 CEST] <durandal_1707> use minterpolate or vs and its mvtools?
[17:16:28 CEST] <Mavrik> Well sure.
[17:16:30 CEST] <frendShot> I'm like "wat", and that's when my boss dropped it on me
[17:16:41 CEST] <Mavrik> But your company sold them that and I'm guessing they paid for it.
[17:16:53 CEST] <Mavrik> So now you either need to go to them with your boss and explain you sold something you can't do.
[17:17:09 CEST] <kepstin> this sounds like a problem that needs an interpersonal solution, not a technical solution :/
[17:17:14 CEST] <Mavrik> Or tell your animators that they need to redraw everything because you didn't read the spec. :P\
[17:17:18 CEST] <frendShot> I don't know how the cash flow goes, but yes, the contract has been signed, and we do owe them what the contract says, but because of this massive clusterfuck I have to come up with a way to fix this
[17:17:41 CEST] <frendShot> yeah, I have a feeling I'll just have to tell my animators and artists that
[17:17:51 CEST] <frendShot> they'll crucify me more than likely, but oh well
[17:18:28 CEST] <Mavrik> I get your shitty situation though :/
[17:18:37 CEST] <Mavrik> But I'm afraid you have a business not a techical problem :(
[17:18:42 CEST] <Mavrik> A common one, but still :(
[17:19:02 CEST] <frendShot> yeah, and a massive interpersonal problem between the producer, and the art director, the latter being myself
[17:19:16 CEST] <frendShot> it's amusing I saw the contract from the ad agency, and not her
[17:20:22 CEST] <Mavrik> We've all been there where sales sold crap that tech can't deliver. :P
[17:21:19 CEST] <frendShot> yeah :/
[17:25:35 CEST] <frendShot> so... with ffmpeg's interpolation filter, there really isn't a way, right? I was trying it out with other animations first to see how it handles similar art styles... and in this animation here, similar to the anime we do, https://youtu.be/zA9zwpMj_8A?t=31s  the busy part with the trail of dust behind Sonic, when I interpolated this to 60fps, that part looks like utter garbage
[17:26:21 CEST] <JEEB> yea, motion interpolation is not magic
[17:26:36 CEST] <kepstin> frendShot: lets put it this way - if motion interpolation did good inbetween animation frames, animation would be a lot cheaper ;)
[17:26:48 CEST] <frendShot> yeah, so I see. I mean, there's other parts of this animation that aren't really busy or highly dynamic where it looks fine and smooth
[17:27:03 CEST] <frendShot> JEEB, kepstin so I see
[17:27:24 CEST] <JEEB> of course you can try tweaking the values
[17:27:32 CEST] <JEEB> of the minterpolate filter
[17:27:46 CEST] <frendShot> i suppose the parts that are chaotic and busy, their issue is that the interpolated parts aren't what would naturally be there, artistically
[17:27:50 CEST] <frendShot> which is why it looks bad
[17:27:52 CEST] <kepstin> yeah, anything with a simple pan would probably look quite decent, a frame with minimal redraw would probably look ok, and anything dynamic, especially with smear frames and whatnot, probably pretty bad.
[17:28:11 CEST] <JEEB> but not much beyond that unless you want to do research and development
[17:28:17 CEST] <frendShot> heh
[17:28:21 CEST] <kepstin> smear frames of course simulate the motion blur that you see when doing low framerate video recording
[17:28:27 CEST] <kepstin> so, yeah :/
[17:28:30 CEST] <frendShot> yeah, exactly
[17:28:36 CEST] <frendShot> so interpolating that would look unnatural
[17:33:44 CEST] <durandal_1707> just blend frames, use framerate filter?
[17:33:52 CEST] <frendShot> hmmm
[18:25:45 CEST] <jsollie> hello everyone, is there someone here with experience of VAAPI? When transcoding movies (tried different muxers, both software and hardware decoding), the transcoding with h264_vaapi works without a problem, but the pixel format is not written in the h264 header.  Can I force this? because this makes the stream unusable
[18:27:35 CEST] <BtbN> h264 doesn't really have a pixel format
[18:31:29 CEST] <jsollie> ok, then what is the 'vaapi_vld' in here:
[18:31:34 CEST] <jsollie> vaapi_vld(progressive, left)
[18:33:20 CEST] <jsollie> Stream #0:0: Video: h264 (h264_vaapi) (High), 1 reference frame, vaapi_vld(progressive, left), 720x576 [SAR 64:45 DAR 16:9], q=0-31, 25 fps, 90k tbn, 25 tbc
[18:33:42 CEST] <jsollie> the vaapi_vld turns to 'none', making it unplayable
[18:33:50 CEST] <jsollie> (in the ffprobe)
[18:34:09 CEST] <kepstin> that's unrelated to why it's unplayable. what's the actual error you're seeing from a player?
[18:35:10 CEST] <jsollie> @kepstin: 'separate color planes not supported'
[18:35:38 CEST] <kepstin> that's a problem with the player/decoder, not with the video itself...
[18:35:56 CEST] <jsollie> ffprobe reports this
[18:36:10 CEST] <kepstin> can you pastebin the *complete* output, please?
[18:36:29 CEST] <jsollie> encoding or decoding output?
[18:36:38 CEST] <kepstin> both, why not.
[18:37:09 CEST] <kepstin> (for decoding output, run a command like "ffmpeg -i video.mp4 -f null -")
[18:39:05 CEST] <kepstin> include the exact command lines you're running, too.
[18:40:03 CEST] <kepstin> also, what hardware are you using? given that it's vaapi, i'd assume AMD or Intel?
[18:41:20 CEST] <jsollie> https://pastebin.com/TCJZm2Ty <-- output
[18:41:36 CEST] <jsollie> sorry, encoding output
[18:41:50 CEST] <jsollie> the GPU is in the log
[18:43:36 CEST] <shuei> hello. this is more of a codec related question rather than ffmpeg. I was using youtube dl to download a lecture and I saw the 1080p version had webm and mp4 link (both video only). mp4 is 377 MB but webm is  715 MB
[18:43:43 CEST] <kepstin> jsollie: alright, other than the fact that amd's hardware encoders aren't great, that encoding log all seems fine.
[18:43:55 CEST] <shuei> discrepancy
[18:43:59 CEST] <shuei> why this discrepancy
[18:44:16 CEST] <kepstin> shuei: I can encode a video with any codec at whatever size/bitrate I'd like by using different options. That's pretty meaningless.
[18:44:21 CEST] <frendShot> shuei, different encoding used
[18:44:28 CEST] <shuei> its on youtube though
[18:44:30 CEST] <frendShot> shuei, download both and see which one looks better; chances are both will look identical
[18:44:36 CEST] <jsollie> @kepstin: ffprobe output of bAndb.m2ts:
[18:44:39 CEST] <jsollie> https://pastebin.com/r4hCTHRR
[18:44:44 CEST] <shuei> wouldn't they use vp9 to reduce size?
[18:44:46 CEST] <frendShot> shuei, yes, YouTube uploads their videos on their servers in possibly multiple encodings
[18:45:03 CEST] <frendShot> shuei, vp9 is hilariously slow to encode, though
[18:45:15 CEST] <shuei> i get that but didn't they move to vp9 to reduce size from mp4
[18:45:43 CEST] <frendShot> I think a bunch of videos coexist in both formats, and I think there's other factors at play there, like how popular the video is
[18:46:30 CEST] <shuei> not very popular but its 2 hours long
[18:46:39 CEST] <kepstin> shuei: they say they moved their *highest quality* encodes to vp9, yes. Youtube's choice of encoding settings - and why they choose particular settings for particular videos -isn't really publicly known
[18:47:17 CEST] <shuei> why would they go through the trouble of encoding to vp9 if the file size is double of mp4
[18:47:42 CEST] <frendShot> kepstin, shuei, I did some snooping with youtube-dl a while ago and it seems the more popular stuff also gets a vp9 encode
[18:47:49 CEST] <c_14> maybe it's higher quality
[18:47:54 CEST] <c_14> also there's no licensing fees on it
[18:48:04 CEST] <kepstin> shuei: are you sure it's even vp9 at all? if it's an older, less popular video it might even be vp8
[18:48:14 CEST] <c_14> though I don't know if you pay per view for h264
[18:48:37 CEST] <JEEB> youtube generally has never cared too much about quality
[18:48:40 CEST] <shuei> kepstin: 248          webm       1920x1080  1080p 2671k , vp9, 30fps, video only, 715.83MiB
[18:49:05 CEST] <JEEB> VP8/9 usage for GOOG has mostly been a) a pride thing (it's their format) and b) a convenient reason to drop all bit rates
[18:49:11 CEST] <JEEB> even if the quality is worse than before
[18:49:13 CEST] <shuei> so now I have to download the audio seperately and merge with ffmpeg now?
[18:49:37 CEST] <kepstin> shuei: youtube-dl can merge them for you
[18:49:58 CEST] <kepstin> jsollie: hmm, it looks like you're hitting some bugs in the vaapi encoder or amd hardware encoder, thb. You might look into trying to disable b-frames on the encoder. Or just use software, it'll give better results anyways.
[18:50:34 CEST] <kepstin> jsollie: if you've got an intel cpu, prefer using that for encoding instead of the amd gpu.
[18:50:43 CEST] <kepstin> intel igpu*
[18:50:44 CEST] <jkqxz> jsollie:  Does it work if you write it to some other container?  That output looks like it's completely corrupted the header somehow.
[18:51:11 CEST] <jsollie> kepstin: yes, but the GPU is hardwired disabled
[18:51:17 CEST] <shuei> 52.4097222 kBps that can;t be a proper video bitrate?
[18:51:30 CEST] <jsollie> jkqxz: mp4 gives also no results
[18:51:31 CEST] <shuei> I divided size/time for the mp4 verison
[18:52:02 CEST] <jkqxz> The current AMD VAAPI implementation has the flaw that it can't supply headers out-of-band, so any container which requires those will have problems.
[18:52:12 CEST] <kepstin> shuei: why not? if the video's mostly static - e.g. slides - that might even look decent. Or you could just have an incomplete download.
[18:52:22 CEST] <jkqxz> jsollie:  Try flat H.264 annex B ("-f h264").
[18:52:25 CEST] <shuei> yeah its more of a presentation
[18:52:33 CEST] <jsollie> and I know the software does better, it's just, well, you know, the challenge :)
[18:52:41 CEST] <shuei> the vp9 version is 100 kbps
[18:54:50 CEST] <jsollie> jkqxz: tried, but the ffprobe output is more or less the same
[18:55:10 CEST] <jsollie> the encoding speed becomes twice as fast, though
[18:55:33 CEST] <jsollie> which would be a nice trick if we can make it work :)
[18:56:54 CEST] <jkqxz> I'd try encode separately, in case the decode messes something up (use hwupload instead of VAAPI decode).
[18:57:11 CEST] <jsollie> yes, I also did that
[18:57:18 CEST] <jsollie> decode is ok
[18:57:43 CEST] <jsollie> encode, even with software decode, gives the ffprobe fail
[18:58:34 CEST] <jkqxz> Try adding "-profile:v 578" to the encoder?
[18:59:17 CEST] <jsollie> so no 'high' profile for amdgpus?
[19:00:09 CEST] <jkqxz> IIRC it claims to support it and then fails in bizarre ways on some GPUs.  That might be your problem here.
[19:00:58 CEST] <jsollie> I also have a tonga gpu here.  it performs exactly the same.  guess this is a case of bad luck then
[19:02:17 CEST] <jsollie> but I have a polaris chip in a few days (currently still in use on a windows machine).  I will try with polaris then
[19:02:37 CEST] <jkqxz> Tonga and Fiji have the same video encode block, I think.
[19:02:47 CEST] <jkqxz> Does non-high profile work, then?
[19:02:57 CEST] <jsollie> sec, I'll try
[19:04:06 CEST] <jsollie> yes, .dump -> .h264 works perfectly
[19:04:07 CEST] <jkqxz> Ohyeah, I did add symbolic names for those.  "constrained_baseline" or "main".
[19:04:31 CEST] <jsollie> in constrained baseline, at least
[19:04:44 CEST] <jsollie> in main, and high, it's a no-go
[19:04:56 CEST] <jsollie> even with -bf 0
[19:07:20 CEST] <jkqxz> Fun.  I thought main did work, but that would probably be on Polaris.
[19:08:24 CEST] <jsollie> Should I open I ticket for this so we can follow up when I have my polaris gpu?
[19:09:02 CEST] <jsollie> *a ticket
[19:11:35 CEST] <jsollie> @jkqxz: according to this wiki page, the VCE >= 2.0 (Fiji is 3.0) should be able to encode B-Frames: https://en.wikipedia.org/wiki/Video_Coding_Engine#VCE_2.0
[19:12:31 CEST] <jsollie> so this becomes confusing
[19:13:58 CEST] <jkqxz> If you want?  In the end it's mainly an AMD/Mesa problem if it advertises support for things which it can't actually do.
[19:15:21 CEST] <jsollie> where would you prefer? mesa or ffmpeg?
[19:16:47 CEST] <jkqxz> Wrt B-frames, I believe VCE 2 supports them but VCE 3 does not.
[19:17:10 CEST] <jsollie> ?
[19:18:27 CEST] <jsollie> ok, so a bonaire gpu should do the job (also somewhere in my closet, but the kernel has no SI support currently)
[19:18:43 CEST] <jsollie> or does it have other issues?
[19:19:47 CEST] <jkqxz> Bonaire definitely can do them with the proprietary driver and libraries on Windows.  I don't think they've even tried to implement it for VAAPI, though.
[19:20:27 CEST] <jsollie> windows is not part of my challenge
[19:21:59 CEST] <jsollie> probably you can force something with dx 11 magic, but I'm not interested
[19:22:53 CEST] <jsollie> should I report this as a mesa or as a drm wish?
[19:32:35 CEST] <jkqxz> Don't mind.  It's on my vague list of "things to look at at some point", but almost nobody seems to care so meh.
[19:39:55 CEST] <jsollie> ok, so I should not expect amd gpus to perform any non-578 encoding? I understand why you decide it, but can you tell me what needs to be done for this? maybe I can help (mostly myself, I suppose)
[19:42:07 CEST] <jkqxz> (578 == 66 + (1 << 10) == baseline + constraint_set1_flag == constrained baseline.)
[19:42:53 CEST] <jkqxz> To make it work AMD needs to do something in Mesa around the codec setup, and possibly supply new proprietary firmware for amdgpu.
[19:43:54 CEST] <jsollie> and I suppose only AMD can fix the firmware?
[19:48:07 CEST] <jsollie> anyway, I will take a look at mesa, but can't promise anything
[19:48:22 CEST] <CSRaghunandan> Hello, we are trying to start a project which requires us to connect 4 cameras to a Hub and run Computer vision algorithms on them. We have identified a few processors for our use case, but we aren't sure if they are feasible
[19:48:26 CEST] <jsollie> thanks for the help
[19:48:41 CEST] <CSRaghunandan> and also, which type of cameras and communication bus to use to connect all 4 cameras
[19:48:43 CEST] <jkqxz> Yeah, the AMD firmware is completely closed so it's up to them to do anything with it.
[19:48:55 CEST] <CSRaghunandan> have anyone worked on a project with similar requirements?
[21:30:21 CEST] <poutine> CSRaghunandan, What kind of computer vision algorithms are you using?
[21:32:25 CEST] <CSRaghunandan> Object detection. We haven't started the project yet. We are still looking to hire a CV engineer for the projecct
[21:34:05 CEST] <poutine> You should probably do that before specing out "the processor", I think you'll likely end up using cloud instances and something like AWS kinesis video or kafka with a well designed system. Not sure anything is specifically a ffmpeg question here
[21:37:27 CEST] <poutine> Also "object detection" is an umbrella term which covers a great many different techniques with different methods of scaling/architecture, your requirements ultimately decide what is needed here, like if you wanted as close to realtime as possible, for instance, YOLOv3 is very fast, but might not meet other expectations
[21:37:30 CEST] <CSRaghunandan> Yea, this wasn't a ffmpeg related question. I asked the question hoping someone who has worked on similar projects could help us out
[23:29:46 CEST] <ThugAim> rip durandal_1707
[23:30:17 CEST] <ThugAim> hey guys, just a quick check... has anyone got cineform going in ffmpeg?
[23:30:25 CEST] <ThugAim> rip lenswipe
[23:46:17 CEST] <ThugAim> btw, anyone here does ffmpeg setups for broadcast automation?
[00:00:00 CEST] --- Tue Sep 18 2018