Comparison of Native 4k, Native 1080p, and Bicubic Downsampled images

Comments

Brandigan wrote on 1/29/2018, 12:01 PM

“A better comparison would be to at least try and take the images in the same 1/4 second”

I can understand your frustration with all of this, unfortunately, my time machine is in for repair.

But your sarcasm machine still works? 😉

I'm not frustrated, this is fun. I'm actually editing some 4k>1080 video in between replies.

As you're already apparently semi-convinced by your results, then I guess you don't need to look further, b
ut if you want to get a closer comparison, having the images you are comparing as temporally synchronized as possible would give you a more accurate result.

So, the shortest clips you can, changing modes as fast as you can, without touching the camera, would do it.

Kinvermark wrote on 1/29/2018, 12:16 PM

Couple more images:

Using GH5 / 8 bit / 420 / 29.97p. Leica lens.

Sample1 - HD rendered to "internet" HD.

Sample2 - UHD rendered to "Internet" HD.

Sorry, dog wouldn't stand still so framing is off. Can still see fine fur detail for comparison though.

Brandigan wrote on 1/29/2018, 12:18 PM

solution, more colour information. That's all it is.

Perfect example. Native 1080p and downsampled 1080p are both presumed to be YUV 8 bit 4:2:0, so the "amount of color information" (number of colors) is the same, save for downsampling.losses from downsampling itself. The color information that is decimated in the conversion from REC2020 is nonrecoverable, no physical exceptions. You can't unbake that cake.

First sentence: Nope. They might be presumed to be by you, but that would be wrong. They don't have to be.

Chroma subsampling is what is happening. True, a single 4:2:0 pixels is set for life.

4 x 4:2:0 pixels (at 1/4 the size) of the same subject will all be slightly different.

So, when you combine them, you will not get the same single pixel the camera gave you originally, because it appears that software does a better (slower) job than "Be as quick as possible, I have things to do, data to write etc." that hardware in the camera does, and so you get back something approaching 4:2:1, or maybe close to 4:2:2 for some pixels.

That's why the results are, IMO, "better". More colour variation in trees and leaves and other small noisy areas, No obvious difference where there was none before in flat areas of colour.

Kinvermark wrote on 1/29/2018, 12:21 PM

It's all just theory until tested: EMPIRICAL DATA PLEASE. Show us YOUR camera's output.

Brandigan wrote on 1/29/2018, 12:21 PM

Couple more images:

Using GH5 / 8 bit / 420 / 29.97p. Leica lens.

Sample1 - HD rendered to "internet" HD.

Sorry, dog wouldn't stand still so framing is off. Can still see fine fur detail for comparison though.

Nice try, but focus point moves from nose to eyes, so can't claim finer fur details is because of downsampling.
Got any plants?

Kinvermark wrote on 1/29/2018, 12:39 PM

Yes, I will try that later (going out now). Could you post some too? Please.

Musicvid wrote on 1/29/2018, 12:59 PM

And now, im going to uninstall this new android keyboard. Yeeesh!

john_dennis wrote on 1/29/2018, 1:02 PM

I don't type (or talk) fast so I won't say much about these concepts. I do try to make decisions based on evidence, though, and my own evidence is preferable to me. I have two out of three examples of the concepts being discussed. Unfortunately, I shot FHD and UHD on different days with this camera.

Screenshot in a Vegas Pro UHD project. 

Screenshot in a Vegas Pro FHD project.

Edit: Added screenshots from UHD and FHD projects.

This picture is a screenshot from a Vegas Pro 14 UHD project. The left-most panel is UHD video from a Canon XC15. The center panel was the same frame saved from the same Vegas project as a TIFF still image sequence, resized to 1920x1080 in Photoshop and placed back into the project. I attempted to faithfully set the pan / crop so the same area was displayed for each panel except the right-most panel which had a different zoom so I left it alone.

I would encourage others to post their examples, and peep at pixels until the cows come home.     

bitman wrote on 1/29/2018, 1:15 PM

What about 4k capable camera's such as RX10 mark 2,3 or 4 - those camera's record the full readout from all of the 20.1 million pixels on its 1-inch size CMOS sensor and this is then down-sampled to a 4K resolution (or to HD)– in other words it is over-sampled and no pixel binning is involved... How does this relate to quality? I presume for such camera's if you want your final output in HD it does not matter if you film in HD or film in 4K and down sample in post (unless you need some cropping, and extra headroom for stabilizing). With such camera, if you do not need the latter, you may prefer in filming in HD, as this has it's own advantages (higher framerates, slow motion, smaller file size, less processing, better battery life etc... ).

I also get the impression (I could be wrong) with my camera (as it does a full read-out) that I have more light available recording in HD as opposed to recording in 4K as more light receptors are available in HD recording (which are normally "occupied in 4K") to be combined to the final HD output file in my camera. More light available as you all know is always better to have sharper pictures as you can change your aperture and shutter speed accordingly. My 2 cents.

Last changed by bitman on 1/29/2018, 1:19 PM, changed a total of 1 times.

Current system: VP 18 (build 284), VP 17 (build 452), (uninstalled VP 12,13,14,15, 16 Suite), Vegasaur, Magix Video Pro X (VPX11), Corel VS ultimate 2019, a lot of NEWBLUE plugins, Titler Pro 6, Mercalli 4.0, Respeedr, Vasco Da Gamma 12, VASST stuff, Production Assistent pro3, Boris Continuum 2020, Davinci Resolve Studio 16,...

  • OS: Windows 10 Pro 64, version 2004
  • CPU: i9900K stepping R0 (since October 2019), previously, der8auer i7-8700K (advanced edition), default speed (no overclock), Cooler: Noctua NH-D15s
  • RAM: G.Skill Trident Z 3200C14 DDR4 64GB, XMP set to profile 1 in BIOS
  • Videocard: NVIDEA RTX 2080Ti (Founders edition), NVIDEA studio drivers
  • Monitor: LG 38 inch ultra-wide (21x9) - Resolution: 3840x1600
  • C-drive (games & APPS): Samsung NVMe SSD 2TB 960 pro

  • Current Video source work drive: Samsung NVMe SSD 2T 970 EVO plus

  • Mass Data storage & Backup: WD gold 6TB + WD Yellow 4TB

  • MOBO: Gigabyte Z370 Aorus Gaming 7, BIOS F14
  • PS: Corsair HX1200i, Case: Silverstone fortress 2,
  • Misc: Logitech G915 (replaced G910), Evoluent Vertical Mouse, shuttlePROv2

 

Musicvid wrote on 1/29/2018, 1:16 PM

It's all just theory until tested: EMPIRICAL DATA PLEASE. Show us YOUR camera's output.

Yes it would seem most important to determine if that level of data integration is NOT ALREADY GOING ON inside your 21st century camera, and if so, is somehow inferior to old-school processing from the 1980s. Tell you another secret, i have run the tests, and it doesnt matter a whit.

More show and less tell will serve you well. you'll get it.

Former user wrote on 1/29/2018, 1:24 PM

“So, the shortest clips you can, changing modes as fast as you can, without touching the camera, would do it.”

Since there are already quite a few contributors to this topic, and a previously similar topic, and no one else by now has requested a redo, my test stands.

———————————————————

One other thing that I’m taking, from my own tests, is that what Pana. and other reviewers claim is true, the FHD on the GH5 is really quite good. Generally the improvement by downsampling 4k to 2k is certainly partly attributable to somewhat inferior FHD in some cameras, to start with. Even though I didn’t use a very high render output data rate, the difference is barely noticeable.

Musicvid wrote on 1/29/2018, 1:42 PM

Bitman, ive already got 2 cents.

You are still immersed in your tests of your a7s and fs7 as you promised, are you not?

Feel free to post back once those tests are completed, sound good?

I just cant compete in that arena, sorry.

Musicvid wrote on 1/29/2018, 1:57 PM

And our late inning pinch hitter, john_dennis, hits a triple on the first pitch!

His methodology is worth studying in detail, as it usually yields more information than the bottom line would suggest. So much so, that i am prepared to accept his conclusions at face value.

 

Kinvermark wrote on 1/29/2018, 2:11 PM

“So, the shortest clips you can, changing modes as fast as you can, without touching the camera, would do it.”

Since there are already quite a few contributors to this topic, and a previously similar topic, and no one else by now has requested a redo, my test stands.

———————————————————

One other thing that I’m taking, from my own tests, is that what Pana. and other reviewers claim is true, the FHD on the GH5 is really quite good. Generally the improvement by downsampling 4k to 2k is certainly partly attributable to somewhat inferior FHD in some cameras, to start with. Even though I didn’t use a very high render output data rate, the difference is barely noticeable.


+1. Very glad I got this camera.

bitman wrote on 1/29/2018, 2:12 PM

Bitman, ive already got 2 cents.

You are still immersed in your tests of your a7s and fs7 as you promised, are you not?

Feel free to post back once those tests are completed, sound good?

I just cant compete in that arena, sorry.

@Musicvid I think you got the wrong person, I do not have a a7s nor fs7, the only 4k kit I own is a Sony RX10 iii and a gopro 😉

Current system: VP 18 (build 284), VP 17 (build 452), (uninstalled VP 12,13,14,15, 16 Suite), Vegasaur, Magix Video Pro X (VPX11), Corel VS ultimate 2019, a lot of NEWBLUE plugins, Titler Pro 6, Mercalli 4.0, Respeedr, Vasco Da Gamma 12, VASST stuff, Production Assistent pro3, Boris Continuum 2020, Davinci Resolve Studio 16,...

  • OS: Windows 10 Pro 64, version 2004
  • CPU: i9900K stepping R0 (since October 2019), previously, der8auer i7-8700K (advanced edition), default speed (no overclock), Cooler: Noctua NH-D15s
  • RAM: G.Skill Trident Z 3200C14 DDR4 64GB, XMP set to profile 1 in BIOS
  • Videocard: NVIDEA RTX 2080Ti (Founders edition), NVIDEA studio drivers
  • Monitor: LG 38 inch ultra-wide (21x9) - Resolution: 3840x1600
  • C-drive (games & APPS): Samsung NVMe SSD 2TB 960 pro

  • Current Video source work drive: Samsung NVMe SSD 2T 970 EVO plus

  • Mass Data storage & Backup: WD gold 6TB + WD Yellow 4TB

  • MOBO: Gigabyte Z370 Aorus Gaming 7, BIOS F14
  • PS: Corsair HX1200i, Case: Silverstone fortress 2,
  • Misc: Logitech G915 (replaced G910), Evoluent Vertical Mouse, shuttlePROv2

 

Musicvid wrote on 1/29/2018, 2:14 PM

Couple more images:

Using GH5 / 8 bit / 420 / 29.97p. Leica lens.

Sample1 - HD rendered to "internet" HD.

Sample2 - UHD rendered to "Internet" HD.

Sorry, dog wouldn't stand still so framing is off. Can still see fine fur detail for comparison though.

Kinvermark. That pooch looks a lot like an old forum celebrity. ☺️

[Edit] Mistaken identity [/Edit]

Brandigan wrote on 1/29/2018, 4:24 PM

It's all just theory until tested: EMPIRICAL DATA PLEASE. Show us YOUR camera's output.


So, you'd trust that I hadn't "tampered" with my images? 😆😆😆😆

No, has to be your own, and better examples than have been posted so far.

But, for those that don't mind doing some actual reading, I found this short article. There are more complicated ones around the interwebs, but seems some people aren't interested in looking too far, so this is really short and to the point.


https://www.eoshd.com/2014/02/discovery-4k-8bit-420-panasonic-gh4-converts-1080p-10bit-444/

 

Musicvid wrote on 1/29/2018, 4:54 PM

Brandigan, I implore you to take voluntary furlough from this thread until you are confident enough to show us some pictures -- your pictures. Or is it really that you are the one with no equipment?

Either way, Strike 2.

Former user wrote on 1/29/2018, 5:24 PM

@Kinvermark

You might find this useful. The main item of interest is the high compression used by some of the GH5 codecs. This is a screen grab of a “numbers” s/sheet I put together shortly after Wolfcrow did his article on the GH5, I based the underlying calculations on his article. Some other cameras are thrown in also.

Musicvid wrote on 1/29/2018, 7:43 PM

Methinks this thread is in better hands without me. Simply outclassed by the amazing talent from guys who have the experience and the newer equipment. Just wow.

fifonik wrote on 1/29/2018, 7:58 PM

It would be nice if someone move this discussion to off-topic.

Camcorder: Panasonic X920

Desktop: MB: MSI B450M MORTAR TITANIUM, CPU: AMD Ryzen 3700X (not OC), RAM: G'Skill 16 GB DDR4@3200 (not OC), Graphics card: MSI RX580 8GB (factory OC), SSD: Samsung 970 Evo+ NVMe 500MB (OS), HDDs: Seagate & Toshiba 2TB, OS: Windows 10 Pro 1909

NLE: Vegas Pro [Edit] 11, 12, 13, 15, 17

Musicvid wrote on 1/29/2018, 8:33 PM

It would be nice if someone move this discussion to off-topic.

Agreed.

Kinvermark wrote on 1/29/2018, 9:25 PM

@Former user thanks for the chart!

@Brandigan Yes, I read that when it was first published. David Newman (Cineoform & Gopro CTO) is a reputable source. But note David is only commenting on chroma subsampling nothing else.

@Musicvid Alas, just the family dog. But I think she's special :) Cheers.

Tim L wrote on 1/29/2018, 9:37 PM

@Musicvid

I'm just a hobbyist trying to follow and understand this thread, but could you explain your original graphic to me?  

By "4K", do we mean 3840 × 2160 (i.e. "UHD-1")?  And these three images (left to right) represent Native 4k, Native 1080p, and 4K downsampled to 1080p?

If so, the leftmost image seems to have four times the resolution in both the horizontal and vertical dimensions, compared to the middle image, rather than two times the resolution in each dimension as I would expect.  But maybe I'm misunderstanding the terminology or misunderstanding the intent of the example.

Thanks for any insight -- I'm just trying to make sense of this.