[Ffmpeg-devel-irc] ffmpeg-devel.log.20191215

[burek]

[00:25:21 CET] <cone-649> ffmpeg 03Marton Balint 07master:28b5dc6199ed: avfilter/vf_interlace: restore lowpass mode constants
[00:25:21 CET] <cone-649> ffmpeg 03Marton Balint 07master:6498522befb2: avfilter/vf_tinterlace: add support for bypassing already interlaced frames
[00:25:21 CET] <cone-649> ffmpeg 03Marton Balint 07master:4cd2cee7ed51: avfilter/vf_interlace: do not interlace already interlaced frames
[00:25:21 CET] <cone-649> ffmpeg 03Marton Balint 07master:1f8e43938bc2: avfilter/x86/vf_interlace: always use unaligned movs
[01:27:17 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:f064c7c449f1: avcodec/mp3_header_decompress_bsf: Check sample_rate_index
[01:27:18 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:c94cb8d9b21b: avcodec/pnm: Check that the header is not truncated
[01:27:19 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:2c2ee67fd36e: avcodec/apedec: Set error on error in ape_decode_value_3860()
[01:27:20 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:185f441ba26a: avcodec/iff: Skip overflowing runs in decode_delta_d()
[01:27:21 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:9d4282658082: avcodec/wmalosslessdec: Check that padding bits is not more than sample bits
[01:27:22 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:ec3fe67074ad: avcodec/wmalosslessdec: Fix several integer issues
[01:27:23 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:5d0b3e282a6a: avcodec/wmavoice: Add FF_CODEC_CAP_INIT_CLEANUP
[01:27:24 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:50db30b47d01: avcodec/adpcm: Fix invalid shift in xa_decode()
[01:27:25 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:a9cc69c0d590: avcodec/wmaprodec: Fixes integer overflow with 32bit samples
[01:27:26 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:6b7bcd437e59: avcodec/decode: Fix integer overflow in max_samples check
[01:27:27 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:92455c8c65c4: avcodec/wmalosslessdec: Fix 2 overflows in mclms
[01:27:28 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:19c41969b26d: avcodec/wmavoice: Check remaining input in parse_packet_header()
[01:27:29 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:a76690c02b4f: avcodec/dstdec: Use get_ur_golomb_jpegls()
[01:27:30 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:5a8877da35bc: tools/target_dec_fuzzer: Adjust threshold for dst
[01:27:31 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:01593278cef0: avcodec/targa: Check colors vs. available space
[01:27:32 CET] <cone-649> ffmpeg 03Michael Niedermayer 07master:9fe07908c3f6: avcodec/adpcm: Clip predictor for APC
[11:44:22 CET] <f00lest> I am using transcoding.c example to get jpeg images out of image frames
[11:45:24 CET] <f00lest> I am doing `avformat_alloc_output_context2(&ofmt_ctx, av_guess_format("image2", NULL, NULL), NULL, filename);` on line 129
[11:45:44 CET] <f00lest> as advised
[11:46:08 CET] <f00lest> but still it encodes the images to h264
[11:46:16 CET] <f00lest> can it be encoded to jpeg instead
[11:46:24 CET] <f00lest> what should be done in that case?
[11:47:28 CET] <f00lest> there is some code about filters, can I just remove this code? https://github.com/FFmpeg/FFmpeg/blob/master/doc/examples/transcoding.c#L375 Is this code useful for my application?
[12:59:33 CET] <MarioMey> JEEB: I'm here again... with the "dagger" in my back, but trying to compile ffmpeg-2.1.5 again. Are you there?
[13:02:07 CET] <BtbN> Why tho
[13:03:31 CET] <MarioMey> I need to compile Blender 2.78... and it uses an old version of ffmpeg. So, I need to compile ffmpeg too.
[13:04:06 CET] <MarioMey> Compiling that old version of Blender in Buster is impossible.
[13:05:00 CET] <MarioMey> I have so much errors... because of ffmpeg.
[13:05:46 CET] <MarioMey> So, I have ffmpeg-2.1.5 to compile, but I don't want to install that version in my O.S. I would like to know how to do it.
[13:06:00 CET] <BtbN> Why not just build a more recent blender version?
[13:06:11 CET] <MarioMey> I need that version.
[13:06:14 CET] <MarioMey> Trust me.
[13:06:33 CET] <durandal_1707> then build without ffmpeg
[13:06:40 CET] <MarioMey> I can't, I need ffmpeg.
[13:07:45 CET] <MarioMey> I use the game engine and I play videos with it.
[13:08:08 CET] <durandal_1707> then there is no solution
[13:27:17 CET] <jdarnley> MarioMey: install that old stuff to some local prefix and it won't interfere with the system
[13:27:43 CET] <jdarnley> Then just tell blender where to find it
[13:27:57 CET] <jdarnley> or rather: blender's build system
[13:29:38 CET] <jdarnley> I don't build such old things but for my dev work I usually use ~/pfx
[13:30:01 CET] <jdarnley> then it is just: ./configure --prefix=$HOME/pfx && make && make install
[15:20:35 CET] <MarioMey> jdarnley, thanks... now I'm compiling in ~/pfx
[15:21:12 CET] <MarioMey> It's funny, before you talk, your name got my attention... I thought it was "darlene"... as in Mr Robot.
[15:21:20 CET] <MarioMey> :D
[15:22:00 CET] <MarioMey> jdarnley!!!
[15:22:25 CET] <MarioMey> I made ./configure --prefix=~/pfx --disable-yasm and it creates a folder with the name "~"
[15:22:39 CET] <MarioMey> lol
[15:24:47 CET] <MarioMey> Sorry, I touched ESC by accident.
[15:26:20 CET] <BradleyS> MarioMey: have you considered spinning up a virtual machine circa the era of your blender version
[15:26:26 CET] <BradleyS> build with older tools
[15:26:36 CET] <BradleyS> like gcc 4.7 or something
[15:31:48 CET] <MarioMey> BradleyS: my intention is to use this Blender version in this O.S.
[15:31:59 CET] <BradleyS> you can
[15:32:20 CET] <BradleyS> but it might save you a lot of grief to build with an older system with those tools
[15:32:52 CET] <MarioMey> The compiled-in-Debian-9 Blender version doesn't work in Debian 10.
[15:33:03 CET] <MarioMey> That's I try to compile now.
[15:33:11 CET] <BradleyS> you would have to build statically
[15:33:31 CET] <MarioMey> Oh...
[15:33:42 CET] <MarioMey> How should I...?
[15:34:29 CET] <BradleyS> assuming it's using a typical configure and makefile system, configure with --disable-shared --enable-static
[15:34:41 CET] <BradleyS> that might be enough
[15:35:11 CET] <MarioMey> Ok, I'll take that in mind...
[15:37:52 CET] <jdarnley> MarioMey: it is the shell that expands ~ and usually only does it when at thet start of an argument
[15:38:26 CET] <jdarnley> So if you want to use it in the middle you need to expand it, or use $HOME
[15:52:22 CET] <MarioMey> Ok, thx.
[20:25:40 CET] <durandal_1707> BradleyS: does app you use to work with braw files contain any info about internal colorspace used (like file info)?
[20:32:17 CET] <philipl> BtbN: did you respond to that guy who was asking about flushing nvenc?
[20:32:53 CET] <BtbN> I don't think I did, where did he ask?
[20:33:07 CET] <philipl> Went to the list on the 18th of november.
[20:33:14 CET] <philipl> "nvenc: Add encoder flush API"
[20:42:46 CET] <BradleyS> durandal_1707: the blackmagic raw player does not, but i'll import into resolve and check there
[20:43:46 CET] <philipl> BtbN: Are encoders allowed to implement flush? avcodec_flush_buffers doesn't state it clearly, although it mentioned decoding.
[20:43:59 CET] <philipl> audiotoolboxenc implements flush
[20:44:18 CET] <philipl> If it's valid, then it's just a matter of implementing .flush for nvenc and people can call it.
[20:50:10 CET] <BradleyS> metadata in resolve only says "black magic raw 24 fps 3840x2160"
[20:50:47 CET] <BradleyS> another thing to consider, there are multiple luts for each camera/sensor they make and each dynamic range preset
[20:51:26 CET] <BradleyS> https://i.imgur.com/8SPRGRq.png
[20:51:42 CET] <BradleyS> so i suspect the curve is dependent on the device's characteristics
[20:51:52 CET] <BradleyS> i only have the pocket 4k
[20:52:21 CET] <BradleyS> (those are 3D LUTs not 1D)
[20:53:03 CET] <BradleyS> but perhaps you are only concerned with the standard curve
[20:57:03 CET] <BradleyS> http://www.bmcuser.com/showthread.php?22311-Oops!-BRAW-isn%92t-Raw-%96-It%92s-a-Y%92CbCr-codec
[20:57:09 CET] <BradleyS> suggests YCbCr, not RGB
[20:58:26 CET] <durandal_1707> its similar but not exactly yuv
[20:58:30 CET] <BradleyS> apparently they're doing debayering in camera (on at least some models)
[20:58:55 CET] <BradleyS> davinci resolve uses an internal editing space they call YRGB
[20:59:26 CET] <BradleyS> a hybrid of yuv and rgb, where you can adjust luminance directly and the rgb color channels directly
[20:59:59 CET] <BradleyS> i wonder if braw is full resolution Y and subsampled RGB instead of UV
[21:01:59 CET] <BradleyS> some screenshots here and analysis might help, but it's mostly speculation: https://forum.blackmagicdesign.com/viewtopic.php?f=2&t=79611&start=0
[21:03:33 CET] <BradleyS> also: https://cml.news/g/cml-raw-log-hdr/topic/braw_codec_analysis/25749037?p=,,,20,0,0,0::recentpostdate%2Fsticky,,,20,2,0,25749037
[21:04:06 CET] <BradleyS> very good info under "What there is"
[21:04:22 CET] <BradleyS> "Transfer curve that is partially linear and partially quadratic, with a threshold"
[21:04:35 CET] <BradleyS> "The DCT compressed data is in a pseudo-YCbCr colour-space, so encoding a weighted RGB average, and differences towards R and B"
[21:04:45 CET] <BradleyS> hope that helps
[21:05:57 CET] <BradleyS> also read "the partial de-mosaic"
[21:11:24 CET] <Lynne> so its applied after yuv conversion?
[21:23:07 CET] <durandal_1707> there is no yuv conversion done
[22:04:18 CET] <durandal_1707> i see some 32768 size arrays being populated by powf(), i wonder is that such transfer function
[22:23:22 CET] <BtbN> philipl, though I'm not sure if nvenc can be flushed
[22:23:32 CET] <BtbN> Without killing it until re-init
[22:31:11 CET] <philipl> BtbN: the email seemed to suggest it can
[22:39:19 CET] <BtbN> I'm still not seeing an E-Mail anywhere. Is it on ffmpeg-devel?
[23:05:42 CET] <durandal_1707> hmm, this color stuff is directly related to what camera uses internally
[23:24:13 CET] <BradleyS> did you see "Linearization LUT, with 32K points" in that last link
[23:24:28 CET] <BradleyS> wonder if that's embedded and camera specific
[23:24:36 CET] <BradleyS> or if that's the sidecar he's writing about
[23:26:51 CET] <durandal_1707> BradleyS: inspecting binary, there is custom luts generated, there is camera specific and other standards ones like 709 and aces, but looks like no camera use that
[23:27:19 CET] <BradleyS> ah, interesting
[23:27:26 CET] <BradleyS> resolve does have aces support
[23:28:17 CET] <BradleyS> not sure why a player would need it if it's not used
[23:28:51 CET] <BradleyS> so do you think the storage format is weighted RGB with offsets?
[23:32:50 CET] <durandal_1707> they basically invented own colorspace and called it Blackmagic Design
[23:32:50 CET] <kierank> durandal_1707: I don't understand
[23:32:56 CET] <kierank> your review
[23:33:08 CET] <kierank> "There is thread_once there."
[23:33:09 CET] <kierank> where?
[23:33:21 CET] <durandal_1707> in init function
[23:33:33 CET] <BradleyS> well, i assume it's intended to play nice with their YRGB space in resolve
[23:33:47 CET] <kierank> durandal_1707: in wrapper I guess?
[23:35:09 CET] <kierank> other codecs use ff_thread_once in init
[23:35:26 CET] <nevcairiel> its litereally there in his init function
[23:35:34 CET] <kierank> oh
[23:35:48 CET] <kierank> i am stupid
[23:36:00 CET] <kierank> sorry
[23:36:14 CET] <durandal_1707> BradleyS: it is marketing thing, for skin tones as they say
[23:41:09 CET] <BradleyS> did this bit in the analysis lend any insights?
[23:41:11 CET] <BradleyS> The partial de-mosaic may just mean:
[23:41:11 CET] <BradleyS>         - take this RGGB raw data, create R, (Gr+Gb)/2, B, convert it to pseudoYUV444, encode with JPEG(444). And the little of Gb-Gr residue is encoded extra
[23:41:11 CET] <BradleyS>         - for fast decoding at 1/2 of resolution just decompress the JPEG and convert to final color space, no de-mosaic, look ma - what a speed!
[23:41:11 CET] <BradleyS>         - for high quality decoding, get the Gr/Gb difference back, compose the original RGGB bayer data and do a full resolution de-mosaic
[23:42:13 CET] <BradleyS> so it's basically jpeg + residuals to create pseudo-raw
[23:42:55 CET] <BradleyS> and color space is RGB->YUV JPEG with residuals to reconstruct the RGB
[23:43:07 CET] <BradleyS> well, reconstruct the RGGB
[23:43:50 CET] <BradleyS> and the color parameters are defined by the 32k point LUT
[23:46:12 CET] <durandal_1707> BradleyS: there is function in their SDK to recreate curve out of parameters
[23:47:01 CET] <durandal_1707> colorspace is neither RGB or YUV. its their own thing
[23:47:34 CET] <durandal_1707> if you have resolve, under metadata you will see color space : blackmagic design
[23:47:56 CET] <BradleyS> ah, i checked in resolve studio 15, not 16
[23:48:44 CET] <durandal_1707> BradleyS: i REd all of it except this last bit that export to standard RGB, you can look at patch on ML
[23:49:00 CET] <BradleyS> ah, did not know there was a patch, will check it out
[23:49:49 CET] <durandal_1707> it uses bayer, but dunno about that, as that is probably not correct will need to think about it
[23:50:03 CET] <BradleyS> does RGB->YUV JPEG + extradata Gb/Gr residuals not qualify as their own color space?
[23:50:53 CET] <durandal_1707> it is not that, it uses own matrix to define colorspace in XYZ domain
[23:51:32 CET] <BradleyS> ah, you mean actual colorspace / matrix, i see
[23:51:36 CET] <durandal_1707> and than store 2 8x8 matrix and 2 4x8 matrix of numbers
[23:52:17 CET] <BradleyS> have you looked at the premiere/avid braw plugins yet?
[23:52:42 CET] <durandal_1707> why? no, i do not have such software
[23:53:00 CET] <BradleyS> just in case the disasm would be helpful
[23:53:20 CET] <BradleyS> they are free
[23:54:04 CET] <BradleyS> i think it's part of the braw sdk, last update today https://www.blackmagicdesign.com/support/
[23:55:57 CET] <durandal_1707> aha
[00:00:00 CET] --- Mon Dec 16 2019