Attila Bakos

I just had a chance to take a look at some X-H2s originals shot in F-Log and F-Log2 in H.265, and it seems to me that the chroma smoothing is way less aggressive in this camera. It's still there, but it won't cause problems I think. This is just my first impression, not a deep analysis. And now you can completely get rid of the smoothing by shooting raw to an external recorder, I tested this with a ProRes RAW HQ clip. What I couldn't test is ProRes internal.
I also noticed that the H.265 files are now tagged with the BT.709 matrix coefficients instead of the BT.601. This will definitely solve some interpretation issues. One small downside is that the files are no longer full range (at least the F-Log and F-Log2 originals I received), they only use the limited range, so we lost some precision here.
 

I meant the reason for the color loss, which is the chroma smoothing. The reasons for the chroma smoothing at this extent I don't know and as you said, we may never know.
I wouldn't say this is a compression issue. This is not something caused by compression algorithms, this looks like some kind of smoothing, tupp already shared some thoughts about that in this thread:
https://www.eoshd.com/comments/topic/60753-color-detail-issues-in-fujifilm-video-files/?do=findComment&comment=471363
 

In my video there is a section where I create a 100Mbps H.265 file from a 3584 x 1730 5D3 raw source and compare it with the 400Mbps Fujifilm file. While it's slightly less pixels than Fuji's UHD, the 100Mbps one is close to indistinguishable from the raw and the Cr channel is very detailed. And I didn't use some well optimized x265 encoder, I created the 100Mbps file in Resolve with GPU acceleration. When you put this next to Fuji's recording which has 4x the bitrate it becomes clear that there must be some purpose to the chroma smoothing other than preserving bandwidth for the luma channel. And I also mention in the video that I checked many files from other manufacturers (some of them have the same amount of luma detail and less bitrate) and didn't see this kind of chroma processing.

I found something strange when I was looking at the chromaticity channels of Fujifilm video files and did some search online, and to my surprise noone is talking about this, so I decided to make a video about it. If you're a Fuji user, you might be interested.
 

Okay, for those who find this kind of stuff interesting, here is a comparison of Cr channels (with added contrast for easy visualisation) from the C70, X-T3, and X-H2s:
https://www.dropbox.com/s/pbfjdx4c2zxhhzq/chroma_compare.mp4?dl=1
C70 is C-Log3, 100Mbps
X-T3 is F-Log, 400Mbps
X-H2s is F-Log, 360Mbps
There is definitely an improvement over the X-T3.
(I only received this files from someone, so I can't do other comparisons.)

I just had a chance to take a look at some X-H2s originals shot in F-Log and F-Log2 in H.265, and it seems to me that the chroma smoothing is way less aggressive in this camera. It's still there, but it won't cause problems I think. This is just my first impression, not a deep analysis. And now you can completely get rid of the smoothing by shooting raw to an external recorder, I tested this with a ProRes RAW HQ clip. What I couldn't test is ProRes internal.
I also noticed that the H.265 files are now tagged with the BT.709 matrix coefficients instead of the BT.601. This will definitely solve some interpretation issues. One small downside is that the files are no longer full range (at least the F-Log and F-Log2 originals I received), they only use the limited range, so we lost some precision here.
 

Okay, for those who find this kind of stuff interesting, here is a comparison of Cr channels (with added contrast for easy visualisation) from the C70, X-T3, and X-H2s:
https://www.dropbox.com/s/pbfjdx4c2zxhhzq/chroma_compare.mp4?dl=1
C70 is C-Log3, 100Mbps
X-T3 is F-Log, 400Mbps
X-H2s is F-Log, 360Mbps
There is definitely an improvement over the X-T3.
(I only received this files from someone, so I can't do other comparisons.)

I just had a chance to take a look at some X-H2s originals shot in F-Log and F-Log2 in H.265, and it seems to me that the chroma smoothing is way less aggressive in this camera. It's still there, but it won't cause problems I think. This is just my first impression, not a deep analysis. And now you can completely get rid of the smoothing by shooting raw to an external recorder, I tested this with a ProRes RAW HQ clip. What I couldn't test is ProRes internal.
I also noticed that the H.265 files are now tagged with the BT.709 matrix coefficients instead of the BT.601. This will definitely solve some interpretation issues. One small downside is that the files are no longer full range (at least the F-Log and F-Log2 originals I received), they only use the limited range, so we lost some precision here.
 

I just had a chance to take a look at some X-H2s originals shot in F-Log and F-Log2 in H.265, and it seems to me that the chroma smoothing is way less aggressive in this camera. It's still there, but it won't cause problems I think. This is just my first impression, not a deep analysis. And now you can completely get rid of the smoothing by shooting raw to an external recorder, I tested this with a ProRes RAW HQ clip. What I couldn't test is ProRes internal.
I also noticed that the H.265 files are now tagged with the BT.709 matrix coefficients instead of the BT.601. This will definitely solve some interpretation issues. One small downside is that the files are no longer full range (at least the F-Log and F-Log2 originals I received), they only use the limited range, so we lost some precision here.
 

I have an EOS R so it will come eventually. I'm also saving up for an R5C and then I can do something for C-Log3 as well.
 
Yes the packs are tailored for specific cameras, I take about 10k color samples from each camera and from these measurements I create LUTs using my own engine. It's quite a long process.

Ah I see. To create Fujifilm Film Simulation LUTs I need to have the camera for 1-2 weeks, and I don't know a single person in my area who owns it. The other thing is that it takes quite a lot of time for me to create these packs so I mainly go for cameras with large user base. In this regard I'm not sure about the F3.

New ACES IDT packs are released:
Leica L-Log/BT.2020 Fujifilm F-Log2/F-Gamut You can get them from here: https://colorizer.net/dev/index.php?op=aces
The Fuji one I did almost blindly, I couldn't get my hands on original F-Log2 clips yet, but since the data sheet is already out I created this package anyway. I used youtube framegrabs for verification and I can already see that if you bring F-Log and F-Log2 to the same contrast, the latter has more saturation, especially in the yellow-red area. And greens are a bit more yellow, foliage seems to be warmer.

But that's already there in Resolve:

 

New ACES IDT packs are released:
Leica L-Log/BT.2020 Fujifilm F-Log2/F-Gamut You can get them from here: https://colorizer.net/dev/index.php?op=aces
The Fuji one I did almost blindly, I couldn't get my hands on original F-Log2 clips yet, but since the data sheet is already out I created this package anyway. I used youtube framegrabs for verification and I can already see that if you bring F-Log and F-Log2 to the same contrast, the latter has more saturation, especially in the yellow-red area. And greens are a bit more yellow, foliage seems to be warmer.

Hello,
This is the official topic for all Colorizer products.
My main focus is bringing high-precision Fujifilm Film Simulation LUTs to multiple cameras.
The following formats are supported at the moment, click the links for videos and details:
Fujifilm X-T3/X-T4 F-Log
Sony S-Log2/S-Gamut3.cine
Panasonic V-Log L & HLG
I also sell ACES IDTs for cameras that are not natively supported by Davinci Resolve:
- Fujifilm F-Log/F-Gamut
- Nikon N-Log/BT.2020
- Sony S-Log2/S-Gamut3.cine
- Z Cam ZLog2
- GoPro Protune
These packs include a lot of 3D LUTs simulating the ACES Rec.709 look, so it can be valuable for Premiere users as well.
I also have Technical fixes for common color shift issues caused by bad interpretation of certain clips in Premiere. I strongly suggest you check this out if you're a Fujifilm user.
As a bonus I created a bunch of free online tools for creating and modifying 3D LUTs, so if you're into this kind of stuff, visit http://colorizer.net.
Thank you for your attention,
Attila

New ACES IDT packs are released:
Leica L-Log/BT.2020 Fujifilm F-Log2/F-Gamut You can get them from here: https://colorizer.net/dev/index.php?op=aces
The Fuji one I did almost blindly, I couldn't get my hands on original F-Log2 clips yet, but since the data sheet is already out I created this package anyway. I used youtube framegrabs for verification and I can already see that if you bring F-Log and F-Log2 to the same contrast, the latter has more saturation, especially in the yellow-red area. And greens are a bit more yellow, foliage seems to be warmer.

And nobody here said it's not, yourself included. A little explanation would have helped, as always.

@tupp nice analysis, I really have to take a look at Darktable again sometime 🙂

It appears that Fuji is essentially using wavelet/frequency-separation in the chroma channels to isolate and blur certain areas.
I tried that scenario on your "5D3 Raw" chroma channel image (captured from your Resolve viewer with MPV), using the Darktable Contrast Equalizer module:

I didn't get it to look exactly the same as the X-T3 chroma channel image, but the "blotchiness" achieved is fairly similar.
Here is the curve used in the Contrast Equalizer module:

The scale from left to right delineates coarse patterns to fine patterns in the image.  The mostly flat line going through the center is the sharpness curve, and the "S-shaped" line is the smoothing curve.
Many use the Contrast Equalizer for denoising and sharpening, but in such applications, the smoothing curve always lies below the sharpening curve.  Nobody ever inverts the two curves as shown on the right side of the graph.
I also used the Soften module with a parametric mask to add extra blur to the brighter areas, because the Contrast Equalizer did not provide enough blur.
The excessive chroma blurring from the X-T3 way too much for demosaicing, and it's extreme overkill for noise reduction.  There also seems to a bit of coarse pixelation revealed when you zoom-in in your video, so perhaps Fuji is additionally doing a pseudo chroma subsampling.
Whatever Fuji is doing (and for whatever reason), the process is effectively reducing the color resolution, which results in a perceptible loss of color depth.
 
The X-T3 is excessively lowering the effective chroma resolution with blurring, which dramatically reduces the color depth.
The raw files don't have that problem, because their resolution is not being reduced.
The "processing" is likely something similar to what I have shown above, and it essentially is reducing the color resolution, hence the lower color depth.  Remember, with digital imaging:
COLOR DEPTH = RESOLUTION x BIT DEPTH
By the way, when are you going to take back that lie you told about me?

I used ffmpeg first to extract the chroma channels into grayscale videos and I used that to validate the results of my plugin. They were identical so it's safe to say that it's not an NLE/GPU decoder issue.

There must be some clever algorithm behind this as it seems to have proper 1080p resolution around well defined edges, but everywhere else it's clearly less than that.

Not at all! Do you plan to include info from the Petapixel article as well, or link that article? With the information there my video makes more sense.

You're not alone with that thought, which is why many including myself left the system. The option to turn off chroma NR should definitely be there, especially since unlike in RAW stills there is no getting around it in video.. 
Like I've been repeating, this an infamous issue in the Fuji community. It's kind of a taboo subject though. Whenever it gets brought up,  people usually dismiss it by saying shoot RAW. I still think its quite jarring as the jpg film simulations are a huge part of the Fuji-X philosophy. 
Again, for videographers there is simply no workaround this issue, hence this topic should really be brought to Fuji's attention imo.
 

Great... I'd rather have an OLPF and slightly less overall resolution than slightly more luma and noticeably less chroma. Or at least they could allow us to modify/switch off the chroma NR.

Unfortunately I won't have the time to test this anytime soon, and I'm not sure it's worth it. The Petapixel artice is absolutely correct about the interpolation issues of the X-Trans pattern, technically there is more guesswork there, and the camera won't have the processing power for a sophisticated demosaicing algo. It's pretty slow even on PC. So I'm sure that the in-camera demosaicing leaves some color artifacts that Fujifilm chose to hide with chroma NR, but they might have gone overkill with that. And I'm surprised that I don't see an improvement on the GFX bodies.