Jump to content

Down-sampled 4k to 2k image vs 2k shot on a 4k sensor


silvan schnelli

Recommended Posts

I've read that down-sampling an image from 4k to 2k can yield benefits in the sense of:

 

  • A 4:2:2 4k image down-sampled to 2k, would result in a theoretical 4:4:4 2k image, leading to a chroma resolution similar to the luma resolution.
  • By binning the pixels, you average random variations (reducing noise) and boost non-random variations (signal), leading to an increase in the SNR
  • It also increases sharpness, although I am not sure how.

Now these points all make relatively sense to me and I'm sure that it also depends on the algorithms used for the down-sampling. However, my question is how does this down-sampled 4:2:2 4k image compare to a 4:4:4 2k image, will it have better chroma resolution at lower storage, will it have less noise, will it be sharper, or is the down-sampled 4k image just better than a 4:2:2 2k image but the same as a 4:4:4 2k image. How much does the algorithm play a role?

And ultimately, if I were to film something would I be better off down-sampling 4k to 2k, or should I just film 2k.

I guess this leads me to another question, when I film 2k on a native 4k sensor and use the whole sensor, does it theoretically bin the pixels automatically, so doing filming 4k and then down-sampling in post would be redundant?

Edited by silvan schnelli
Link to comment
Share on other sites

  • Premium Member

2k is a dead format, been dead for around a decade now. 

Where it's true, the Alexa 35 is still only a 4k camera, that is the lowest acceptable delivery resolution today for many reasons. 

1) Bandwidth (bit rate) is king. A 1080p (2K) distribution, will have WAY less bandwidth than a UHD (4K) distribution. You always want to distribute in 4k because the more bandwidth the better it will look. The difference between 4:2:0 (standard home streaming) 4:2:2 and 4:4:4 on someone's home TV is unnoticeable in 4k, but in 1080p, eh you can see if it you have some hard color lines between the red and blue channel. 

2) Future proofing your production is critical. Delivering in 12 bit 4:4:4 4k (Pro Res 4444 or XQ) is the only real way to deal with this. 

3) Most distributors will want the production to have a minimal of 4k acquisition. This isn't a Netflix BS standard, it's just the way things go. I had a film rejected by QC recently because we had some scenes shot in 2k within a 4k finished film. We had to fight with the distributor. Where it's not "necessary" for distribution, most people will request it. 

I generally try to shoot digitally in 6k or 8k so that when you deliver in 4k, you've got a full 444 sample and if you shoot raw, 12 bit finish is at your fingertips. That's really the way things are done in the modern world, even with motion picture film. Everyone scans at much higher res and down samples to 4k. 

  • Like 1
  • Upvote 1
Link to comment
Share on other sites

@Tyler PurcellThank you for the very informative answer on perhaps the most important question, addressing in which resolution I should film in.

I will definitely have to do more research into bandwidth and how that ties into resolution. I am surprised to hear that 2k is a dead format, as i thought that  many movies were still mastered in 2k, or do you really just mean 2k acquisition?

Nevertheless, I would also still like to know the answer to some of my other questions, regarding the effectiveness of Downsampling.  

Link to comment
Share on other sites

its not uncommon still for VFX to be done at 2k, especially if there is an overwhelming volume to deal with, and then those shots get upscaled. But that is almost always taking at least a 4k image at the beginning of the pipeline. Last I checked 2k and 4k DCPs are still limited to the same bit rate too. BUT 1080p blurays are limited to 8bit, where as 4k blurays are 10bit. In a scenario where you had a 10bit or 12 bit 2k image, I'd upscale the master to 4k before sending to a disc printer if you want that 10bit goodness.

As for what happens if you shoot 4k, downsample to 2k, then upscale back to 4k, Im not sure what color benefits you'd get. I suspect the tools in davinci could give you an idea if you tested it though. I've still got a sony F35 (5k sensor resolving to 1080) which I use with ultra primes, to my eye its 4k upscale is perfectly fine. But thats more a fun stuff camera not a client camera.

What is the shooting scenario that prompted this question? is it theoretical or are you planning something? those details may help us give better advice, though I do agree with Tyler that if you have the option to shoot natively 4k thats kinda expected now. Even with 16mm, I only scan at 4k or above

  • Upvote 1
Link to comment
Share on other sites

@Robin Phillips Ah I see very interesting. I am not to familiar yet with bit rates and especially how they are tied to resolution. Intuitively I assumed that higher resolution footage required higher bitrate and lower resolution footage required a lower bitrate, but that due to this they would ultimately playback at the same quality. However what I seem to be getting from your answers is that this isn't the case. I suppose I still don't understand the bitrate is king statement, are there any links I can refer to, to learn more about this.

Now regarding the scenario that prompted the question, well there are two reasons. One is that we have a film exercise where the professors want us to have the final output in 2k and wanted to know if it makes sense to film 4k and then down-sample to improve SNR (besides the obvious benefit of being able to crop at 4k).

So it would be important for me to know how much down-sampling from lets say ProRes 422 at 4k to 2k differs, to filming ProRes 444 at 2k on a 4k sensor (with respect to all the things I mentioned in my original post).

I am also planning to shoot many tests in which I'd like to explore and discuss all the different methods in which the SNR can be improved when filming and down-sampling would be one I'd like to tackle in this experiment.

Link to comment
Share on other sites

bitrates are most important in color reproduction, but its possible to have a super low bitrate 4k (or even 8k) image or a very high bit rate 2k image. I'm one of those people who has long argued a 1080p 444 stream is gonna be way better than a 4k 420 stream, but the world went with resolution first. easier to sell TVs that way and all I suppose. But if you've got a controlled color environment and you say had a 10 bit 444 1080 stream that you could side by side compare with an 8 bit 420 stream of the same footage, the 10 bit stream would pop way more. You're talking an order of magnitude change in the amount of color by going to 10bit and above.

If your delivery spec is 2k, and you know your camera's 2k mode will use the whole sensor to create a 2k image and not crop, theres nothing wrong with that. if you're really interested in this though, I'd suggest getting a color chart and shooting it with a variety of settings so you can pull it into resolve and compare. If youre shooting with anything remotely modern I wouldnt worry too much about the signal to noise ratio, but Im also someone who has no qualms about degraining 16mm to cheat for 35 in some circumstances.

But if you dont have time with the camera to experiment extensively, I'd just shoot in 4k and do a 2k (or 1080, whatever is asked for) output from resolve. 

  • Upvote 1
Link to comment
Share on other sites

@Robin PhillipsAh okay yeah I see that makes sense to me. However, that arises a lot of question from me  in regards to Tyler Purcells first point. (Not saying he is wrong just that I am confused)

1) Bandwidth (bit rate) is king. A 1080p (2K) distribution, will have WAY less bandwidth than a UHD (4K) distribution. You always want to distribute in 4k because the more bandwidth the better it will look. The difference between 4:2:0 (standard home streaming) 4:2:2 and 4:4:4 on someone's home TV is unnoticeable in 4k, but in 1080p, eh you can see if it you have some hard color lines between the red and blue channel. 

Since you need higher bandwidth for 4K and innately less for HD, as you mention, why don’t we just user a higher bandwidth for HD, wouldn’t that improve the quality then. I am just confused cause the argument is to shoot 4k for more bandwidth but it also requires more bandwidth and at constant bandwidth a 1080 would look better than 4k. I’m sure I’m misunderstanding something here, or do we just use the minimum amount of bandwidth required for streaming at each resolutions, well in that case shouldn’t it just all be the same cause 4k needs more cause it’s bigger so it would playback at the same quality as 1080 which needs less cause it’s smaller?

 

And yeah I think I’ll definitely have to run some of my own tests to see what the noise and quality difference will be like in regards to my original question of downsampling. 

 

thank you for your time and replies.

Edited by silvan schnelli
Link to comment
Share on other sites

its not inherent, Tyler is referencing standard distribution specs for 2k and 4k. You or I can make a 1080p render at 12 bit and the maximum bitrate the software will allow, but thats not gonna pass QC for delivery. Fine for a master, but your distributor will have hard specs you have to hit for different formats (steaming, bluray, SD, etc). 1080p blurays max out at 8 bit color and 40Mbps I think. 4K bluray can triple that with 10 bit color. 

We dont do this universally because of old standards that we have to stick to. Notably youtube premium added a higher bitrate option to their non 4k downsamples as an experiment in quality. I also havent done hard testing on this, but for a while it sure looked like disney + had better color reproduction than netflix. 

  • Upvote 1
Link to comment
Share on other sites

  • Premium Member

FYI; I'm referencing streaming, not Pro Res final delivery. 

Pro Res HQ for 2k vs 4k will have compensatory bandwidth for the frame size. 

.h264/.h265 will not. 

So where it's true, if you're going to theatrical DCP, none of this is an issue. The vast, vast, vast, vast majority of people are going to "self" distribute their production. They will need to do the final compression and delivery. Which is the framework of my discussion. 

Yes, 2k is entirely dead. Nobody is finishing in 2k anymore. That ended in around 2019? Sure, some lower-end productions will do visual effects shots in 2k due to time/cost. The vast majority of theatrical shows, even streaming, will be 100% 4k today. I work with VFX houses quite a bit and they do a few cool tricks within the composites where they can output 4k, but have multiple resolution sources within the composite to save considerable time. Once they got that process down, from that time on, it's been nearly entirely 4k for the top shows. 

 

  • Like 2
  • Upvote 1
Link to comment
Share on other sites

On 10/16/2023 at 3:19 PM, silvan schnelli said:

@Robin Phillips Ah I see very interesting. I am not to familiar yet with bit rates and especially how they are tied to resolution. Intuitively I assumed that higher resolution footage required higher bitrate and lower resolution footage required a lower bitrate, but that due to this they would ultimately playback at the same quality. However what I seem to be getting from your answers is that this isn't the case. I suppose I still don't understand the bitrate is king statement, are there any links I can refer to, to learn more about this.

Now regarding the scenario that prompted the question, well there are two reasons. One is that we have a film exercise where the professors want us to have the final output in 2k and wanted to know if it makes sense to film 4k and then down-sample to improve SNR (besides the obvious benefit of being able to crop at 4k).

So it would be important for me to know how much down-sampling from lets say ProRes 422 at 4k to 2k differs, to filming ProRes 444 at 2k on a 4k sensor (with respect to all the things I mentioned in my original post).

I am also planning to shoot many tests in which I'd like to explore and discuss all the different methods in which the SNR can be improved when filming and down-sampling would be one I'd like to tackle in this experiment.

 

That is your best bet...test.

See theory in practice. Share your tests with us when you get them done. 

  • Upvote 1
Link to comment
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now
×
×
  • Create New...