10 bit RGB DPX file identified as rgb48le

[dave rice]

These DPX image files as 10 bit RGB; however ffmpeg identifies them as rgb48le rather than gbrp10le.

Sample here: ​http://www.archive.org/download/Sample10BitRgbDpxFile/00255.dpx

ffmpeg -i 00255.dpx
ffmpeg version N-35294-g2d38081 Copyright (c) 2000-2012 the FFmpeg developers
  built on Mar 12 2012 21:13:33 with gcc 4.2.1 (Apple Inc. build 5666) (dot 3)
  configuration: --prefix=/usr/local/Cellar/ffmpeg/HEAD --enable-shared --enable-gpl --enable-version3 --enable-nonfree --enable-hardcoded-tables --enable-libfreetype --cc=/usr/bin/gcc-4.2 --enable-libx264 --enable-libfaac --enable-libmp3lame --enable-libtheora --enable-libvorbis --enable-libvpx --enable-libxvid --enable-libopencore-amrnb --enable-libopencore-amrwb --enable-libass
  libavutil      51. 42.100 / 51. 42.100
  libavcodec     54. 10.100 / 54. 10.100
  libavformat    54.  2.100 / 54.  2.100
  libavdevice    53.  4.100 / 53.  4.100
  libavfilter     2. 64.101 /  2. 64.101
  libswscale      2.  1.100 /  2.  1.100
  libswresample   0.  7.100 /  0.  7.100
  libpostproc    52.  0.100 / 52.  0.100
Input #0, image2, from '00255.dpx':
  Duration: 00:00:00.04, start: 0.000000, bitrate: N/A
    Stream #0:0: Video: dpx, rgb48le, 2048x1556 [SAR -1:-1 DAR 512:389], 25 tbr, 25 tbn, 25 tbc
At least one output file must be specified

[Carl Eugen Hoyos]

The dpx file is packed RGB and has 10 bits per component and is correctly identified (and transcoded).
At the time the decoder was written, the only colourspace that could save information was RGB48.
This could now be changed to planar GBR10, but I don't think it makes much sense to convert from packed to planar, so I will close this ticket.

[dave rice]

When I convert the DPX to raw rgb48le then there is random data in bits 11-16 for each channel. Is the conversion of this DPX to raw rgb48le not a lossless conversion? Where does this extra data come from?
...
OK, looking closer it seems that bits 11-16 are a copy of bits 1-6, so the 10 bit DPX sample decodes to bit:1,2,3,4,5,6,7,8,9,10,1,2,3,4,5,6. Is there a way to prevent this, as it seems to just add noise in the lesser significant bits?

[Michael Niedermayer]

Replying to dericed:

When I convert the DPX to raw rgb48le then there is random data in bits 11-16 for each channel. Is the conversion of this DPX to raw rgb48le not a lossless conversion? Where does this extra data come from?
...
OK, looking closer it seems that bits 11-16 are a copy of bits 1-6, so the 10 bit DPX sample decodes to bit:1,2,3,4,5,6,7,8,9,10,1,2,3,4,5,6. Is there a way to prevent this, as it seems to just add noise in the lesser significant bits?

The bits replication is needed to maintain pure white and pure black. Consider 1bit mono to be converted to lets say 8 bit without replication it would become gray/black.