4K to 1080P 4:4:4

D7K wrote on 10/10/2016, 12:14 PM

Have you seen this:


I did a test with 4K from my Gx8 and it appears to be much better than rendering to any of the 4:2:0 codecs.


Here is a quote from the OP at DPR:



I guess that depends on whether you're a GH3/GH4 user or an Oly user, I certainly think that if Panny stays a part of the forum and they don't split (which IMHO is sensible) then video questions are fine, although a [video] tag (at the start of the subject) like the one suggested for Equivalence might enable people who don't have an interest to skip them.

Oh and thanks for pointing out there's a new Cliptoolz, although with a different name, I thought they had stopped developing it.

As to the original question then scaling a 4k image to 2k (i.e. by 2x in H and V and 4x overall) will certainly increase the information content. Remember if you average four values (e.g. with 7,9,9,10 you get 8.75) you get two extra "bits" of information (from resolution of 1 to 0.25, although not as perfectly if did it at the analogue level, they might have started off as 7.4, 9.2, 8.75, 10.3 -> 8.91 -> 9 at 0.25 resolution).

For colour 4:2:0 gives half the chroma resolution in both x and y and we're halving the x and y resolution. So you go from a chroma sample per 4 (2x2) pixels to one per pixel. 4:4:4 is one per pixel. Think of it this way:
4:4:x means every 4 horizontal luminance pixels you get four chroma pixels (Cr, Cb, Cr, Cb)
4:4:4 means you get different samples for each chroma value on the scan line below (i.e. another set of Cr, Cb, Cr, Cb).

4:2:x means in a set of 4 luminance pixels the first two pixels have a Cr and the second two a Cb
4:2:0 means the next line uses the luminance pixels from the line above and no has no more chrominance information.

So if we do a 2x2 scaling on the luminance pixels but keep all the chrominance pixels then over each two lines of 4 pixels (made from 8 lines of 4) we get an output pixel with alternating Cr, Cb values for every line, which is 4:4:4.

E.g. scaling 8x4 to 4x2

4:4:4 -

Cr Cb Cr Cb Cr Cb Cr Cb
Cr Cb Cr Cb Cr Cb Cr Cb
Cr Cb Cr Cb Cr Cb Cr Cb
Cr Cb Cr Cb Cr Cb Cr Cb


Cr -- Cb -- Cr -- Cb --
-- -- -- -- -- -- -- --
Cr -- Cb -- Cr -- Cb --
-- -- -- -- -- -- -- --

4:4:4 native at half H and V resolution -

Cr Cb Cr Cb
Cr Cb Cr Cb

(Note if scaled higher res to get here would have multiple chrominance samples per output sample)

4:2:0 scaled by 2 in H and V -

Cr Cb Cr Cb
Cr Cb Cr Cb

(Note each chrominance sample is just copied from the higher res image.)

Sorry for issues due to the font, may be clearer cut and pasted into Notepad or similar."



Kinvermark wrote on 10/10/2016, 12:44 PM

No comment about the sub-sampling math, but anecdotally I agree that 4k footage downsampled to HD looks great!   Also, Gopro's David Newman is a seriously knowledgeable source.

It is not always easy to see a difference between 420 footage and 422  (depends on the content) or 8 bit vs 10 bit.

I use a GH4 and feel   that the final product of 4k camera footage is better than the final product of its 1080p modes.


AVsupport wrote on 10/10/2016, 3:39 PM

I own a A6300 and one reason I purchased was the advertised 4K from the full APSC sensor without pixel binning. Now it turns out the 1080 is relatively soft, ergo binning somewhere, surprisingly, and shooting in 4K overheats quickly. HDMI is 8bit only so there's no real advantage in extra clutter. I don't understand why Sony don't manage to properly downscale to 1080 when others know what to do. what a shame.

my current Win10/64 system (latest drivers, water cooled) :

Intel Coffee Lake i5 Hexacore (unlocked, but not overclocked) 4.0 GHz on Z370 chipset board,

32GB (4x8GB Corsair Dual Channel DDR4-2133) XMP-3000 RAM,

Intel 600series 512GB M.2 SSD system drive running Win10/64 home automatic driver updates,

Crucial BX500 1TB EDIT 3D NAND SATA 2.5-inch SSD

2x 4TB 7200RPM NAS HGST data drive,

Intel HD630 iGPU - currently disabled in Bios,

nVidia GTX1060 6GB, always on latest [creator] drivers. nVidia HW acceleration enabled.

main screen 4K/50p 1ms scaled @175%, second screen 1920x1080/50p 1ms.

Kinvermark wrote on 10/10/2016, 4:32 PM

You mean 1080 direct from camera is soft?   Overheating aside, is the 4k from camera downrezzed in post better?