2K vs 4K shooting comparison at DVinfo.net

David Newman · April 9th, 2007, 12:01 PM

If preparing for NAB we are creating timelines with real-time editing of 4K, 2K, 1080p, and other HD formats, I thought it would be interesting to see what happens when you combine them and what are the visual quality advantages. We can do this mixing in real-time on a Intel dual core (I have even done limited 4K real-time on my old 2.8Ghz P4 -- so imagine what a modern PC can do.) Please visit our NAB booth (SL7826 -- near Apple but down the stairs) to check it out.

My write up and testing is on my blog : http://cineform.blogspot.com

But you can place you discussion/comments here (better than the blogger comments field.)

Bob Grant · April 11th, 2007, 06:50 AM

I have to agree, all else being equal doubling resolution doesn't produce much bang for the buck. The difficulty though is in keeping 'all else equal'.
If we can get images from the SI-2K as good as a 2K DI from 35mm we'll be extremely happy but I suspect that's not going to be that easy or cheap once glass is factored in.

Jim Arthurs · April 11th, 2007, 10:09 AM

Looking forward to FINALLY meeting you in person, David... somehow missed you on the last three NAB's.

I've always been a big proponent (as a RED reservation holder) of shooting in 4K and finishing at 2K/1080. Virtually every other post of mine at RedUser makes some mention of this in one form or another. My reasons range in difficulty providing quantities of CGI at 4K, the value of re-framing and undistorting lens issues, etc. Plus the non-trivial fact that all projection outlets are currently not 4K. There isn't even a direct pixel-to-pixel way to view your footage on a monitor running at speed.

My feeling is that the 4K RED will be the best possible 2K camera, whether sampled down in camera, or later in post, which is how I will treat it. A 4K bayer sensor downsampled to 2K "wins" over a native 2K bayer sensor, and you don't even need to look at a test to know this, it's common sense.

Having agreed with you so far, I will have to argue that your test isn't really that useful until you can compare apples to apples with sensor output from 4K and 2K windowed from the RED to itself, and compare both of these it to any other 2K native sensor. Every native 2K bayer still I've seen is soft, soft, soft! Where if you start with 4K and sample down, you'll have natural (not edge enhanced) sharpness, better color space, and as I said earlier, the "perfect" 2K end product.

Now, if the test was flipped, and you COULD monitor 4K easily, allowing you to scale UP the 2K version and compare to the native 4K on a 4K timeline, I think it would be very apparent how much more the 4K brings to the table.

BTW, here's a grab from a 2K and 4K sample from the "milk girls" images posted at RED. The are from slighly dissimilar moments in time, as they never released the same frame in both resolutions, but still a useful comparison. I then resized the 2K up to match the 4K dimensions.

http://ftp.datausa.com/imageshoppe/o.../2K_vrs_4K.png

Regards,

Jim Arthurs

David Newman · April 11th, 2007, 10:29 AM

I agree the 4k wins over 2K, but agrument is based on the delivery medium than the audience will not notice. Shoot in either 2K or 4K, post in 2K, save time and space, editing in RT on commodity Intel hardware (you can see this at our NAB booth SL7826.) I do believe the test I did is valid, as generating 2K and 4K frames of the same image, FOV & DOF is near impossible, with the same camera or two camera, so the simulation is the best we can do. The only question is whether the OLPF strength is a valid simultation, I welcome all feedback on that one.

Jim Arthurs · April 11th, 2007, 10:43 AM

David, I still don't understand why you need to blur in the first place... doesn't the Dalsa have an OLPF already applied to the 4K image you're working with? I know the RED will.

Adding additional blur as part of the downsample process seems biased against the value of the 4K source. You say without blurring, the result is aliased... how can that be? Isn't it just your viewing process or display method in Premiere that's aliased? Any time I downsample a larger image to a smaller one, the effective sharpness (natural sharpness) is increased. The process of downsampling alone, with good downsampling, is enough to control any potential aliasing issues. Only in interlaced displays do you need to do some EXTRA vertical blurring to keep twitter in control.

I agree that we should aim FINAL resolution to the display target, 2K or 1080p. I'm just saying that NO 2K native Bayer sensor will give you that "every pixel unique" end result that a 4K Bayer downsampled to 2K will. Every single 2K Bayer image I've seen is SOFT! You need 4K to get good 2K. You need greater than 4K to give perfect 4K for that matter... but we're a way away from having to deal with that...

Regards,

Jim Arthurs

Jason Rodriguez · April 11th, 2007, 11:02 AM

If I'm understanding correctly, the 4K source was blurred further to derive the 2K RAW source (simulating the effect of the OLPF in front of the 2K sensor). The original 4K RAW source (without filtering), and the derrived 2K RAW source (with a simulated OLPF "blur") were then put on the same 2K timeline (so the 4K is automatically downsampled to 2K without a blur filter applied) and compared.

Jim Arthurs · April 11th, 2007, 11:08 AM

Yes, but Jason, WHY would you want to do that if the goal is the best possible 2K image you can output? That's just simulating a less than full raster 2K image when you COULD have a full raster one by directly downsampling and skipping the blur...

This soulds too much like the various methods the prosumer cameras do when claiming HD... and I know David would agree with me on that... the HVX for instance... we both agree that it doesn't fill 1280 by 1080 when producing 1080, and the "best" way to handle the HVX output it to downsample to full raster 1280 by 720... you wouldn't blur when doing that!

Regards,

Jim A.

Jim Arthurs · April 11th, 2007, 11:17 AM

Just re-read your post Jason... if you're correct on the steps David took, then ignore what I wrote...

Jim A.

David Newman · April 11th, 2007, 11:31 AM

Making it clear (restating Jason's correct comment), the 4K image scaled to 2K was not blurred (it is the full quality), only the 2K RAW simulation was blurred during its downscale. Blurring is necessary to simulate the OLPF of a 2K camera, as without it you get aliased images like the noX camera.

Rob Johnson · July 24th, 2007, 07:32 PM

I was doing quite a bit of reading again today regarding the debate between 2K vs. 4K. I'm all for progess and a better picture, to put it in layman's terms. I wonder, unless you're looking to wow your audience from the side of the Hindenberg, I wonder if it's too much for typical viewing environments. Why not just settle the debate once and for all and introduce the 7K IMAXCam? 70 million pixels would get you a really, really nice picture (said partially with tongue in cheek). Or should we just consider this the inevitable "HD" revolution for theater? Any opinions?

April 9th, 2007, 12:01 PM	#1
David Newman CTO, CineForm Inc. Join Date: Jul 2003 Location: Cardiff-by-the-Sea, California Posts: 8,095	2K vs 4K shooting comparison If preparing for NAB we are creating timelines with real-time editing of 4K, 2K, 1080p, and other HD formats, I thought it would be interesting to see what happens when you combine them and what are the visual quality advantages. We can do this mixing in real-time on a Intel dual core (I have even done limited 4K real-time on my old 2.8Ghz P4 -- so imagine what a modern PC can do.) Please visit our NAB booth (SL7826 -- near Apple but down the stairs) to check it out. My write up and testing is on my blog : http://cineform.blogspot.com But you can place you discussion/comments here (better than the blogger comments field.) __________________ David Newman -- web: www.gopro.com blog: cineform.blogspot.com -- twitter: twitter.com/David_Newman

April 11th, 2007, 10:29 AM	#4
David Newman CTO, CineForm Inc. Join Date: Jul 2003 Location: Cardiff-by-the-Sea, California Posts: 8,095	I agree the 4k wins over 2K, but agrument is based on the delivery medium than the audience will not notice. Shoot in either 2K or 4K, post in 2K, save time and space, editing in RT on commodity Intel hardware (you can see this at our NAB booth SL7826.) I do believe the test I did is valid, as generating 2K and 4K frames of the same image, FOV & DOF is near impossible, with the same camera or two camera, so the simulation is the best we can do. The only question is whether the OLPF strength is a valid simultation, I welcome all feedback on that one. __________________ David Newman -- web: www.gopro.com blog: cineform.blogspot.com -- twitter: twitter.com/David_Newman

April 11th, 2007, 11:31 AM	#9
David Newman CTO, CineForm Inc. Join Date: Jul 2003 Location: Cardiff-by-the-Sea, California Posts: 8,095	Making it clear (restating Jason's correct comment), the 4K image scaled to 2K was not blurred (it is the full quality), only the 2K RAW simulation was blurred during its downscale. Blurring is necessary to simulate the OLPF of a 2K camera, as without it you get aliased images like the noX camera. __________________ David Newman -- web: www.gopro.com blog: cineform.blogspot.com -- twitter: twitter.com/David_Newman

April 11th, 2007, 06:50 AM	#2
Bob Grant Trustee Join Date: Nov 2005 Location: Sydney Australia Posts: 1,570	I have to agree, all else being equal doubling resolution doesn't produce much bang for the buck. The difficulty though is in keeping 'all else equal'. If we can get images from the SI-2K as good as a 2K DI from 35mm we'll be extremely happy but I suspect that's not going to be that easy or cheap once glass is factored in.

April 11th, 2007, 10:09 AM	#3
Jim Arthurs Regular Crew Join Date: Sep 2004 Location: Colorado Springs CO Posts: 120	Looking forward to FINALLY meeting you in person, David... somehow missed you on the last three NAB's. I've always been a big proponent (as a RED reservation holder) of shooting in 4K and finishing at 2K/1080. Virtually every other post of mine at RedUser makes some mention of this in one form or another. My reasons range in difficulty providing quantities of CGI at 4K, the value of re-framing and undistorting lens issues, etc. Plus the non-trivial fact that all projection outlets are currently not 4K. There isn't even a direct pixel-to-pixel way to view your footage on a monitor running at speed. My feeling is that the 4K RED will be the best possible 2K camera, whether sampled down in camera, or later in post, which is how I will treat it. A 4K bayer sensor downsampled to 2K "wins" over a native 2K bayer sensor, and you don't even need to look at a test to know this, it's common sense. Having agreed with you so far, I will have to argue that your test isn't really that useful until you can compare apples to apples with sensor output from 4K and 2K windowed from the RED to itself, and compare both of these it to any other 2K native sensor. Every native 2K bayer still I've seen is soft, soft, soft! Where if you start with 4K and sample down, you'll have natural (not edge enhanced) sharpness, better color space, and as I said earlier, the "perfect" 2K end product. Now, if the test was flipped, and you COULD monitor 4K easily, allowing you to scale UP the 2K version and compare to the native 4K on a 4K timeline, I think it would be very apparent how much more the 4K brings to the table. BTW, here's a grab from a 2K and 4K sample from the "milk girls" images posted at RED. The are from slighly dissimilar moments in time, as they never released the same frame in both resolutions, but still a useful comparison. I then resized the 2K up to match the 4K dimensions. http://ftp.datausa.com/imageshoppe/o.../2K_vrs_4K.png Regards, Jim Arthurs

April 11th, 2007, 10:43 AM	#5
Jim Arthurs Regular Crew Join Date: Sep 2004 Location: Colorado Springs CO Posts: 120	David, I still don't understand why you need to blur in the first place... doesn't the Dalsa have an OLPF already applied to the 4K image you're working with? I know the RED will. Adding additional blur as part of the downsample process seems biased against the value of the 4K source. You say without blurring, the result is aliased... how can that be? Isn't it just your viewing process or display method in Premiere that's aliased? Any time I downsample a larger image to a smaller one, the effective sharpness (natural sharpness) is increased. The process of downsampling alone, with good downsampling, is enough to control any potential aliasing issues. Only in interlaced displays do you need to do some EXTRA vertical blurring to keep twitter in control. I agree that we should aim FINAL resolution to the display target, 2K or 1080p. I'm just saying that NO 2K native Bayer sensor will give you that "every pixel unique" end result that a 4K Bayer downsampled to 2K will. Every single 2K Bayer image I've seen is SOFT! You need 4K to get good 2K. You need greater than 4K to give perfect 4K for that matter... but we're a way away from having to deal with that... Regards, Jim Arthurs

April 11th, 2007, 11:02 AM	#6
Jason Rodriguez Trustee Join Date: Mar 2003 Location: Virginia Beach, VA Posts: 1,095	If I'm understanding correctly, the 4K source was blurred further to derive the 2K RAW source (simulating the effect of the OLPF in front of the 2K sensor). The original 4K RAW source (without filtering), and the derrived 2K RAW source (with a simulated OLPF "blur") were then put on the same 2K timeline (so the 4K is automatically downsampled to 2K without a blur filter applied) and compared.

April 11th, 2007, 11:08 AM	#7
Jim Arthurs Regular Crew Join Date: Sep 2004 Location: Colorado Springs CO Posts: 120	Yes, but Jason, WHY would you want to do that if the goal is the best possible 2K image you can output? That's just simulating a less than full raster 2K image when you COULD have a full raster one by directly downsampling and skipping the blur... This soulds too much like the various methods the prosumer cameras do when claiming HD... and I know David would agree with me on that... the HVX for instance... we both agree that it doesn't fill 1280 by 1080 when producing 1080, and the "best" way to handle the HVX output it to downsample to full raster 1280 by 720... you wouldn't blur when doing that! Regards, Jim A.

April 11th, 2007, 11:17 AM	#8
Jim Arthurs Regular Crew Join Date: Sep 2004 Location: Colorado Springs CO Posts: 120	Just re-read your post Jason... if you're correct on the steps David took, then ignore what I wrote... Jim A.

July 24th, 2007, 07:32 PM	#10
Rob Johnson Regular Crew Join Date: Jul 2007 Location: Ft Myers, Florida Posts: 65	I was doing quite a bit of reading again today regarding the debate between 2K vs. 4K. I'm all for progess and a better picture, to put it in layman's terms. I wonder, unless you're looking to wow your audience from the side of the Hindenberg, I wonder if it's too much for typical viewing environments. Why not just settle the debate once and for all and introduce the 7K IMAXCam? 70 million pixels would get you a really, really nice picture (said partially with tongue in cheek). Or should we just consider this the inevitable "HD" revolution for theater? Any opinions?