search

felx / ffmbc

Automatically exported from code.google.com/p/ffmbc
0 stars 0 forks source link

RGB DPX frame to Prores 422/444 HQ produces colour shift to yellow #114

Open GoogleCodeExporter opened 8 years ago

GoogleCodeExporter commented 8 years ago 
What steps will reproduce the problem?
1. DPX frame with full RGB values to Prores HQ 422 or 444 encode with ffmbc 
command
2.
3.

What is the expected output? What do you see instead?
I wouldnt expect any colour shift from RGB to YUV or from DPX RGB to Prores RGB

What version of the product are you using? On what operating system?
FFmbc 0.7rc7 for Windows (64-bit) 5-Jul-2012
Windows 7 64bit

Please provide any additional information below:

C:\Users\User\Desktop\ffmpeg-PRORES_ENCODING>ffmbc -f image2 -r 25 -i 
JUNK_DPX_TEST\%06d.dpx -vcodec prores -profile hq N:\ffmpeg_encodes\output.mov
FFmbc version 0.7-rc7
Copyright (c) 2008-2012 Baptiste Coudurier and the FFmpeg developers
Input #0, image2, from 'JUNK_DPX_TEST\%06d.dpx':
  Duration: 00:00:40.04, start: 0.000000, bitrate: N/A
    Stream #0.0(und): Video: dpx, rgb48le, 1920x1080p [PAR 65536:65536 DAR 16:9], 25.00 fps
File 'N:\ffmpeg_encodes\output.mov' already exists. Overwrite ? [y/N] y
Incompatible pixel format 'rgb48le' for codec 'prores', auto-selecting format 
'yuv422p10le'
[scale @ 0000000001AFD200] w:1920 h:1080 fmt:rgb48le -> w:1920 h:1080 
fmt:yuv422p10le flags:0x4
Output #0, mov, to 'N:\ffmpeg_encodes\output.mov':
  Metadata:
    encoder: FFmbc 0.7
    Stream #0.0(und): Video: prores, yuv422p10le, 1920x1080p [PAR 1:1 DAR 16:9], 183500 kb/s, 25.00 fps
Stream mapping:
  Stream #0.0 -> #0.0
Press [q] to stop, [?] for help
frame=  256 fps= 19 q=1.0 size=   10534kB time=00:00:10.24 
bitrate=8427.5kbits/s eta=00:00:39.63

Original issue reported on code.google.com by tho...@thelooklondon.com on 21 Aug 2012 at 12:02

GoogleCodeExporter commented 8 years ago 
This actually seems to be effecting a conversion from DPX to rawvideo 422 QT as 
well

Original comment by tho...@thelooklondon.com on 21 Aug 2012 at 2:59

GoogleCodeExporter commented 8 years ago 
I will need the original dpx to reproduce

Original comment by baptiste...@gmail.com on 1 Sep 2012 at 1:56

GoogleCodeExporter commented 8 years ago 
Here you go Baptiste.  If you convert this frame to a QT and then compare the 
colour you'll see the output gets slighter warmer.  Have tried adding -vf 
colormatrix=bt709:bt709 with no change.

Original comment by tho...@thelooklondon.com on 11 Sep 2012 at 4:40

Attachments:

GoogleCodeExporter commented 8 years ago 
Hummm, djv and imagemagick cannot convert this dpx, their output is a black 
frame.
What are you using to view and create the dpx ?

Original comment by baptiste...@gmail.com on 11 Sep 2012 at 6:07

GoogleCodeExporter commented 8 years ago 
That was from a Quantel Pablo, this one is from SGO Mistika, be aware the frame 
is very dark

Original comment by tho...@thelooklondon.com on 12 Sep 2012 at 2:51

Attachments:

GoogleCodeExporter commented 8 years ago 
Having now also done a Uncompressed YUV422 QT to Prores I also get the same 
shift.  I think this is the same as this issue: 
http://web.archiveorange.com/archive/v/DUtyPyCWLv29zJqkZkPl
Maybe as we're in a colour critical environment we pick up on this when others 
dont care, however this means ffmpeg can not be used for true 1:1 color 
critical work.  A snapshot of our scope shows a change in the RGB parade (the 
last three).  The Red and Blue channel is different, unfortunately i cant 
upload these as this forum says the storage quota for this thread is exceeded.  
Thanks

Original comment by tho...@thelooklondon.com on 14 Sep 2012 at 10:01

GoogleCodeExporter commented 8 years ago 
Hey guys, I can confirm that we at Weta are experiencing the same issue. I have 
the latest ffmpeg and ffmbc versions compiled and i've ran a few tests encoding 
a dpx sequence into a prores quicktime. We're trying to find a replacement for 
our current in-house solution that runs only on mac os x and uses native 
quicktime libraries (we are finding the code maintenance to be very difficult, 
which is why we are wanting to find an alternative) and i can generate an 
awesome looking prores movie, but as OP posted, it is more yellow than the 
source. There is a definite color shift. The quality of the video and encoding 
is better (prores HQ, 48fps, 2k res, yuv44410p, 170Mbit data rate) and the 
codec we use ATM (apple intermediate), but the color is off. 

I'd be happy to provide some dpx's and resulting sequences generated with our 
in house solution vs. ffmpeg for comparison if this could help solve the 
problem...

thanks
jakub

Original comment by jakub.kr...@gmail.com on 23 Oct 2012 at 5:20

GoogleCodeExporter commented 8 years ago 
Humm, the other frame looks plain black to me, I cannot distinguish anything.
I'll close this issue unless a good dpx file is supplied that clearly 
illustrates the problem,
I just cannot fix anything without a good test case

Original comment by baptiste...@gmail.com on 1 Nov 2012 at 8:28

GoogleCodeExporter commented 8 years ago 
I have noticed some of the same symptoms. If the limit of attachments hadn't 
been full I'd attached an archive with two dpx'es, two quicktimes, and two 
screenshots from video scopes in FCP. I know it's not the most accurate, but 
our hardware video scope confirms the shift.

I'll try emailing you the archive :)

Commands used to produce quicktimes:

ffmbc -i calibrate.%04d.dpx -r 25 -vcodec prores -profile std -vf 
colormatrix=bt601:bt709 -y dpx_to_prores_709_filtering.mov

ffmbc -i calibrate.%04d.dpx -r 25 -vcodec prores -profile std -y 
dpx_to_prores_no_filtering.mov

Original comment by flehnerh...@gmail.com on 1 Nov 2012 at 10:18

GoogleCodeExporter commented 8 years ago 
I've uploaded a kodak marcie(rec709) dpx and its converted to prores 422 
output, and a split screen as a tiff Baptiste. I've shared the files with you 
via your gmail account.
Hopefully this will clearly show you the issue.  Its clear that the guys at 
Weta have the same problem and they know what they're doing!
Best regards

Original comment by tho...@thelooklondon.com on 22 Nov 2012 at 6:17

GoogleCodeExporter commented 8 years ago 
Has this topic now been closed?

Original comment by tho...@thelooklondon.com on 7 Jan 2013 at 10:24

GoogleCodeExporter commented 8 years ago 
The issue still hasn't be fixed as of FFmbc-0.7-rc8 or FFmpeg 2.1.1

Original comment by nicolas....@gmail.com on 6 Jan 2014 at 5:29

GoogleCodeExporter commented 8 years ago 
disappointing that this has never been fixed, would be great to use a command 
line render farm on windows to do these! oh well

Original comment by tho...@thelooklondon.com on 27 Jan 2014 at 4:16

GoogleCodeExporter commented 8 years ago 
It would be a game changer...

Original comment by stevem...@gmail.com on 27 Jan 2014 at 5:41

GoogleCodeExporter commented 8 years ago 
Has anyone tried the colormatrix filter? IIRC there is a 601 default baked into 
the code somewhere.

As far as I can see, only Rec.709 (sRGB primaries), FCC, Rec.601 (ITU-R 
BT.470-2/SMPTE 170M), and SMPTE 240M are supported, but it would appear they 
should work for the src / dst conversions.

"-vf colormatrix=src:dst" where src / dst are one of:
‘bt709’
‘bt601’
‘smpte240m’
‘fcc’

Example:

"-vf colormatrix=bt601:bt709"

If someone could test this and report back, that would be excellent.

Original comment by troy.sob...@gmail.com on 2 Feb 2014 at 10:49

GoogleCodeExporter commented 8 years ago 
I tried this a year ago, i still had an issue.  would be interested to hear
if anyone else had more success

*I'm often with clients and unavailable for fast email replies, so for
urgent bookings and enquiries please call us, or email: *
bookings@thelooklondon.com* and you will get a response*.
*Book early to avoid disappointment*.

*THE LOOK*
29-35 Rathbone Street
London, W1T 1NJ
T: 0207 2875313
E: thomas@thelooklondon.com
www.thelooklondon.com
2013: Voted 5th in Televisual's Top Colourists survey
2012: Voted 2nd in the 'Craft' category in Broadcast magazine's 'Hot 100
People in the industry'
Go to our website to see our latest drama, features and commercials work

Original comment by tho...@thelooklondon.com on 2 Feb 2014 at 10:56

GoogleCodeExporter commented 8 years ago 
Is there any way someone following this thread can do a test with the Bella 
Nuit (http://www.belle-nuit.com/test-chart) test pattern and sample the luma 
across the two 709 / 601 sets?

I have a strong suspicion that RGB to YCbCr conversion is using the 601 
matrices. This would account for the filter effects being skipped perhaps.

Original comment by troy.sob...@gmail.com on 5 Feb 2014 at 4:31

GoogleCodeExporter commented 8 years ago 
I've identified a number of issues causing this shift (all in libswscale).
First of all, the bt601 coefficients are hard coded in swscale.c (which does 
rgb->yuv). Even when changing those, the luma will still be wrong if mmx is 
enabled (because mmx also assumes bt601 for any yuv material).
swscale.c.rc8patch will change the defult coefficients to bt709 and disable mmx 
for the scaler. 

When converting back from yuv->rgb, bt601 coefficients will also be used.
swscale.h.rc8patch changes yuv->rgb coefficients to bt709.

However, when converting from yuv444 to rgb, rc8 will not give you full color 
resolution.
utils.c.rc8patch fixes this with some lines borrowed from latest ffmpeg.

NB! Even though utils.c.rc8patch seems to do the trick in this case, I might 
have broken something else. If you only need rgb->yuv and not the other way, 
you only need swscale.c.rc8patch

Bengt

Original comment by bengt...@hocusfocus.no on 9 Mar 2014 at 12:46

Attachments:

GoogleCodeExporter commented 8 years ago 
Thanks for this Bengt, unfortunately I've been unable to compile the RC8 with 
these patches for a Win 7 64 machine which is very frustrating (the main source 
of info online is for 32bit).  Does anyone know how to do this and can send me 
the outputted ffmbc.exe for testing Bengt's fix?
Thanks all

Original comment by tho...@thelooklondon.com on 15 Mar 2014 at 11:12

GoogleCodeExporter commented 8 years ago 
Unless ofcourse it is possible to disable MMX in the command line encode in 
RC8?  in which case what would be the correct command for RGB DPX rec709 to 
Prores 422 HQ?

Original comment by tho...@thelooklondon.com on 15 Mar 2014 at 1:34

GoogleCodeExporter commented 8 years ago 
I like the utils.c patch I will apply it.
Other patches seem intrusive to me, if it could be conditional if would be 
better.

Original comment by baptiste...@gmail.com on 1 Jul 2014 at 1:50

GoogleCodeExporter commented 8 years ago 
I would think that full range RGB should only be dumped if the source is full 
range (yuvj)?

Does it dump full range if the source is yuvj range?

Original comment by troy.sob...@gmail.com on 1 Jul 2014 at 3:50