How good is Sony's XAVC-I codec when compared to ProResHQ?

[DougJensen]

Sometimes I hear people say that Sony's XAVC-I codec isn't good enough for shooting LOG and that they prefer recording  Apple's ProRes codec with an external recorder.  Using an external recorder for this reason always seemed like a huge waste of time and effort to me, but I'd never done a comparison between the two.  So I decided to do some tests between XAVC-I and the HQ version of ProRes.

XAVC-I @ 24 fps records at 240 Mbps while ProResHQ is over 700 Mbps. That means that (all other things being equal) a ProResHQ file will be 3x larger than the XAVC-I version. In other words, you can get 3x more footage on a memory card when you shoot XAVC-I. But do you pay a penalty in image quality when using XAVC-I? To answer that question, I decided to shoot a few test shots with my FX6 to see how the two codecs compared. I recorded XAVC-I internally onboard the camera while simultaneously recording the camera's 10-bit output via HDMI to a Blackmagic 5" Video Assist. In these split-screen tests, the XAVC-I version is on one side of the screen . . . and the ProResHQ version is on the other. Both were shot with the exact same camera settings (S-LOG3) and have had the same grading applied in post. Only very minor adjustments have been made in Resolve to fine-tune and match the Lift/Gamma/Gain. Can you tell which is which?

 

 

[Mark Mosrie]

Hey Doug, really appreciate this test. It's really, really close and I thought I might be imagining any difference I perceived. In the end, I felt that the footage on the right — the "B" footage — had slightly better contrast and detail. If I had to guess, I'd say that was the ProRes. Can I convince you to tell me which was which?

Thanks!

Mark Mosrie 
mark@penumbra-ent.com

Edited November 11, 2023 by Mark Mosrie

[DougJensen]

On 11/10/2023 at 10:59 PM, Mark Mosrie said:

 Can I convince you to tell me which was which?

Hi Mark, no problem!  It is actually in the YouTube description.

*** Spoiler Alert ***

Test 1: A: XAVC-I B: ProResHQ

Test 2: A: ProResHQ B: XAVC-I

Test 3: A: ProResHQ B: XAVC-I

Test 4: A: XAVC-I B: ProResHQ

Test 5: A: ProResHQ B: XAVC-I

Test 6: A: XAVC-I B: ProResHQ

[joema]

I have tested simultaneously-recorded S-Log3 FX6 4k/23.98 10-bit 4:2:2 XAVC-I vs 4k/23.98 ProRes 422 (recorded on Atomos) many times. A few people have alleged XAVC-I has significant "macro blocking" issues on certain scene types, especially complex tree branches moving in the wind. The videos they've posted allegedly showing these used zoom factors of 400% to 1000%. 

In general I can't see any difference when viewed normally or even when cropped to the maximum 200% limit for maintaining equivalent resolution for 4k on 1080p presentation.

When comparing them in an NLE using overlayed clips with the top layer using "difference" compositing instead of "normal", you can see some *tiny* differences but when viewed normally, I can't see differences except in rare cases on paused frames at about 400% or so magnification. 

Here are some original camera files of simultaneously-recorded FX6 XAVC-I S-Log3 vs ProRes 422 from a Shogun 7. Also included is a split-screen 200% full-res TIFF file showing the same frame from both formats. It's not labeled, but according to those claiming the XAVC-I problem, it is so obvious that anybody can tell the difference under normal viewing conditions. However, I cannot tell the difference even at 200% magnification -- and of a scene type that supposedly manifests the problem: https://www.dropbox.com/scl/fi/40nkhlin8x6gn3mzy07oo/SHGN7_S001_S001_T021.MOV?rlkey=itzbt86ie81ti8ron7pg9wueo&dl=0

Also, the MXF container used by the FX6 has much better metadata than externally-recorded ProRes. For S-Log3 it enables the "raw controls for XAVC" feature of Resolve Studio 18.5 and later. It's not true RAW, but it enables RAW-like adjustments in post for white balance, tint, exposure, etc.