1DC: 1DX II: Premiere Pro CC 4K 422 Decoding Bug

[jcs]

While PP CC did a good job decoding C300 II RGB 444 and upscaling to 4K (with better GS keys than 422 YUV 10-bit 4K), it's got a bug with 1DX II YUV 422 4K MJPEG:

PP CC 2019 (latest on the PC): 1DX II 4K 422 MJPEG (4096x2160 in a UHD sequence):


Loaded into Resolve (latest) and rendered out as a UHD QuickTime Cineform 10-bit YUV (Wavelet) and brought back into PP CC:


This looks like a 422 color space handling issue (horizontal (only) color problem). Found this bug after noticing strange artifacts after keying.

Colorspace (601/709 bug?) and levels (data/video bug?) also different between them. Transcoding in Media Encoder to QT Cineform lessens the appearance of the bug and keeps the same color (more magenta vs. more green in Resolve).

Is it possible to use Resolve to transcode without changing colorspace/levels?

 

[jcs]

The issue appears to be horizontal pixel swapping: Pixel Dyslexia.

Correct horizontal pixel order:
0 1 2 3 4 5

What appears to be happening:
1 0 3 2 5 4

Reported to Adobe on their forum.

EDIT: upon further review this looks more complicated than simple pixel swapping.

 

[NorBro]

Honestly, not sure anyone knows what you're talking about jcs, lol.

You've just created an entirely new level of pixel-peeping.

 

[jcs]

It's ok man, I've posted on the Adobe forum as well as submitted a bug report. Folks who do keying and care about 420/422/444 for VFX understand these issues. On normal viewing levels, the edges look a little jaggy, so for our (simple) YouTube content, not really a big deal. More disappointing is paying $53 a month for buggy software (that I could fix myself for free).

Workarounds include transcoding to QuickTime Cineform: a neat discovery- really nice quality for file size. Did some experiments and found decent results with smaller files are possible as YUV 10-bit (and fast to edit) vs. MJPEG 8-bit 422 (even with the bump up to 10-bit (which buys nothing unless downsampling 4K to 1080p)). A side effect is the bug is less apparent. Transcoding with Resolve 100% solves the issue, but color/levels change (could probably put together a 3 LUT to fix it (prefer the Premiere color/look vs. how Resolve is working)).