Cineform decoder distorts lower 8 pixels on 1080 video

[Garbord]

Summary of the bug:

The lower 8 pixels of 1080 height video distorts when decoding Cineform video.

This does not occur when tested against 1088 height so may be related to 1080 not being divisible by 16.

This is also visible when replayed in ffplay.

The same source file does not show the same errors when replayed in Windows Media Player, using the installable GoPro Cineform codecs.

I've been able to see this on multiple clips. All generated from Adobe Premier Pro CC.

How to reproduce:

% ffmpeg -i input.mov -vcodec rawvideo -pix_fmt bgr24 output.mov
ffmpeg version N-87327-g18821e3 Copyright (c) 2000-2017 the FFmpeg developers
  built with gcc 7.2.0 (GCC)

[Garbord]

Truncated log of ffmpeg console output

[Garbord]

Uploaded sample file 'Cineform_Bottom_8_Pixel_Distort_1080_YUV.mov' via VLC Uploader.

[Carl Eugen Hoyos]

Unfortunately, I don't see the file there (the upload site may not work at all).
Please use datafilehost or dropbox or another file hosting site of your choice.

[Garbord]

Replying to cehoyos:

Unfortunately, I don't see the file there (the upload site may not work at all).
Please use datafilehost or dropbox or another file hosting site of your choice.

I've uploaded the sample file here:

​https://drive.google.com/open?id=0B9D-zOmjHZlrTUpCRnVldmNhTm8

I've also included a png screenshot and full version of the log.

[Carl Eugen Hoyos]

Thank you for the sample, I agree that it shows a possible issue with FFmpeg.

I am not sure I understand completely though:
On the file host, you provided an output file for comparison (you should not never do that unless asked, this ticket is another good example why), the png image looks very colourful here, and a little ugly on the bottom, as indicated by you.
You also uploaded a cineform file (I wonder why it is so big: Is there any difference between the first frame and the other frames, except for the timestamp?) - it looks less colourful here, actually black-and-white.
Is there any connection between the mov file and the png file you provided? Is the bug that the mov file is supposed to show colours other than black and white?

Just allow me to say that I am not 100% convinced that uploading a 1GB sample with three (or four?) different frames is super-useful.

Not a regression afaict.

[Carl Eugen Hoyos]

Distortions in both lower corners visible

[Carl Eugen Hoyos]

If there a distortions, they are less visible

[Carl Eugen Hoyos]

Sample with the most obvious distortions uploaded:
​http://samples.ffmpeg.org/ffmpeg-bugs/trac/ticket6675/

[Garbord]

Sorry, its the first time Ive submitted a bug. I will keep samples smaller in the future.

[Garbord]

The distortion is also visible with Cineform RGB files, though less obvious.

[Kieran Kunhya]

Probably fixed in d2917501c252d999e5478c4b804e326ac77cf37b

[Garbord]

Thanks, I'll check this against some different export files tomorrow.

Off topic I know, but now that GoPro have open sourced their code are there any plans to integrate an encoder? I was hoping to talk to someone to get an idea of the technical challenges involved, to see if it is something to hope for in the near future. Is there a particular point of contact I can discuss this with?

[Garbord]

Distortion still visible on all submitted sample files (bw1, bw2, colour).

I tested using Zerano Windows build (x64) version N-89212-ga60b2425c3 (​https://ffmpeg.zeranoe.com/builds/win64/static/ffmpeg-20171123-a60b242-win64-static.zip)

[Elon Musk]

Fixed in 673604e0e3b7c7d6a077f61dee322d534ff6eae5.