Welcome to our community

Be a part of something great, join today!

Log in
  • Hey all, just changed over the backend after 15 years I figured time to give it a bit of an update, its probably gonna be a bit weird for most of you and i am sure there is a few bugs to work out but it should kinda work the same as before... hopefully :)

Steve Yedlin, ASC on Resolution

Wrote this in some Facebook thread where they discussed the same thing and all the BM camera owners where so happy that their pocket cameras shooting 2k actually is better or atleast as good as any other camera. :)


Copy pasted from there:

Björn Benckert Saw all of part two now. This is so much fud that I dont know where to start. I could pick any segment in that clip and explain why he is wrong in more ways than one. Sorry.
LikeShow more reactions · Reply · 4 hrs
Manage
Björn Benckert
Björn Benckert So lets start with the easy bits:

Why does alexa 65 look better than the XT on closeup, grain halation and so on on pretty much every test done?... Sure, you can say different sized sensor but except for that. smaller noise better hard contrast edges etc. The camera has the exact same sensor tech as the XT and the Amira and the mini the only thing differ is the pixel count and sensor size. Same pixels but more of them... image looks better in all the different zoom in tests and also in the zoom out tests and also in the noise test, and yes pretty much in every little node tree this guy shows the XL looks better than the XT....

Now alexa 65 is a (i dont remember the exact numbers but it´s about) 6.5k something c-mos camera so in other words it resolve 1/3 to 1/4 of that which is about 2.5k so when looking at the alexa XL in this 4k crop in demo what do you actually see? Well you see at best lets say 3k but it´s really not better than 2.5k upscaled to 4k. If you took a higher resolving camera like a Phase One or such and shot a 18k image and placed it into a scene with a alexa XL in a native XL sized timeline... The phase one would stand out as a razor blade.

So simply this guy is sitting looking at cameras that resolve about 3k at best and say he can not see any difference between them, or little difference between them. Then we sit and look at the same test on a heavily compressed internet feed and say we also do not see any difference at these cameras shooting something like 2k when watching them in 4k....

Also he points at film and calls it 35mm film... he does not state what ASA. on film ASA is very determine to resolving resolution. So comparing film and saying it looks like this is kind of like taking a digital camera shoot a picture and say this is what digital images looks like... does not matter if the image is shot with a iPhone or a Phase one or Alexa XT.... it´s reference for digital camera, thats how those looks. Simply the silver particles on celluloid films vary in size for different ASA. So does the resolution. Another thing looking at film as a still frame is not fare. Film vary the layout of the silver flakes from one frame to another so that actually gives the viewer more info about the scene when watching a clip than just a still frame. In other words film played has a different resolution than a still frame...

Noise, noise is info even if it´s unwanted info it´s info so adding noise increase the amount of information in the frame but possibly not wanted info. Less captured noise is more info of the scene and less random info. Normally downsampling helps reduce noise.

Furter more when comparing the cameras. Exposure is important. it´s obvious that the exposure for each camera in this test might have been the same or at least not fully optimized for the camera in question. As most of these cameras have different DR layouts and also DR widths. That makes huge difference when he zoom in and talk about detail resolution on images where noise is visible. Simply when you see noise in an image that is considered underxposure. Sure that could be done to taste but when comparing the resolution off two cameras it´s possibly good practice not to introduce noise as noise lowers resolution. Compressed noise even more so.

So the grain determines the detail, and when shooting compressed the grain / noise ratio is deteriorating the image in a exponential way. So looking at the cameras and not talking about the difference in DR layouts is simply not right at all. He might as well shoot with the lens cap on with one of the cameras and say look it does not look any different in 4k or 2k or SD.

Downscaling an image from 6k to 4k can not be done with even numbers. It also does not give you an image where the camera is having info for red green and blue for each pixel. It´s also very much important what downscale algoritms is used. It´s also very important if there was an OLPF in play or not and if sharpens is applied before or after downscale (sharpens should normally be applied if OLPF is in use)

The talk in the end about multiple resampling for different mastring formats etc. Sure might apply to some but today it´s quite possible to stay in native resolution and create masters from it and then downsample with even numbers. For example shoot 8k create a UHD master and what ever needed DCP straight from there and then create the HD master from the UHD master. Which is resolution devided by 4... Which is not really that destructive. And if wanted the HD master could also be driven straight from the native source resolution. We work that way and it works well.

Then also not to mention when stablizing images oversampling makes huge difference. Simply its quite difficult to rotate a picture with a fraction without loosing a lot of sharpens. Rotating a 8k frame and setting it down in a 4k frame is far better results than scaling it down to 4k and then rotate it... simple as that, resolution matters.

The theater bit with the shart of the screens etc. I have a 5k LG oled screen for my mac I also have several lower resolution displays of the same size and HD macbook´s and 4k laptops. Now those that can not see the difference when playing a 4k feed on a 4k laptopscreen instead of a hd feed need glasses or you need to put you laptop on your lap, simply you see the difference. I also got a 4k 65" oled screen, and when watching anything HD I dont really care if I sit in the back seat or up front but when watching the netflix Dolby vision 4k stuff... then I really want to move up close, why is that. Why am I seeing such big difference between netflix HD and Netflix UHD even if they are both heavily compressed?

Also Better Call Sal to me it looks horrible in 4k but when looking at my shitty macbook pro in HD it looks quite fine. It´s a dragon shot series but to me they went too high on iso and did pretty shitty post, noise all over the place and focus is not always right, bleading windows and so on. So when watching it in 4k I see all those errors. House of cards season 4 shot with the same camera, the quality when watch in 4k vary but way better overall than The Better call Call stuff.

Now I had an epic for several years, then dragon and now helium. Sure 3 different sensors and 3 completly different cameras, with different bitrates etc. I use them for all sorts of things VFX work, stills and normal filming stuff.
You do not need to be a rocket scientist to see that it´s really like 1,2,3 in picture quality. And yes again more things than the K´s differ but someone would need more than 2 hours to convince me that they are equal when it comes to creating a 4k mastered image. Simply put helium outshines the other two like mad... When creating a HD image well then difference is not as big and for SD I would not really care much which one of the 3, or I would actually take a dragon as that one has larger sensor at higher frame rates.

But sure it´s not only about resolution my 4k DJI osmo´s does for most scnearios look less then on par with the alexa XT even if the alexa has less resolution. It´s one of the few quality parameters in a camera that you can put a number on. Color accuracy noise ratios etc is nearly impossible to pin down what each camera is actually doing. So naturally Netflix, and BBC and others set a bar in resolution and then also add other spec´s that cameras must fulfill to be accepted. Normally there is a list of the most common cameras that are accepted on those lists, and usually those lists makes sense... The 600USD DJI osmo is not on the list just like the BMPCC camera for obvious reasons.
 
Frank Glencairn said:
The main takeaway for me is, that compression kills any resolution. Something I was saying for years. Whats the point of having a 6k+ sensor, when your camera is averaging out boxes of 16x16 pixels at the recording level. That's why I prefer an uncompressed raw out of my BM cameras, over any compressed 6k+ material. We also need better delivery codecs, not more pixels.
Click to expand...

DSMC3 needs NVMe mags with speeds up to 2.000 MB/s (like the Samsung 960 pro) so that the users will have more choice in compression from lossless 1:1 to 1:??
Computers get faster, storage gets bigger and faster so there is no reason to stick to mSATA 300 MB/s.
 
Björn Benckert said:
Sure does extra pixels does not make any difference what so ever. :)

A002_C003_11289I by Björn Benckert, on Flickr
Click to expand...



Isn't this a bit disingenuous? You're comparing sensors as if their size was directly related to their pixel count. With this type of false comparison you could put the Alexa 65 within that frame and it would be much smaller than the Helium 8k. The same 6k 65mm sensor that outperformed all others in this test, thus making the case for much of what Yedlin is arguing.

Also full frame "open gate" on the Alexa is 3414 x 2198.
 
Misha Engel said:
DSMC3 needs NVMe mags with speeds up to 2.000 MB/s (like the Samsung 960 pro) so that the users will have more choice in compression from lossless 1:1 to 1:??
Computers get faster, storage gets bigger and faster so there is no reason to stick to mSATA 300 MB/s.
Click to expand...

Just as an FYI, REDcode uses wavelet based compression. 1:1 is not possible with REDcode. If you want to see "uncompressed" that is to say mathematically uncompressed images from a RED sensor, then shoot some frames using REDcode burst at 2:1 compression. Those shots would be mathematically lossless.

In other words, if it goes through REDcode it is always compressed, but at 2:1 the compression The data is mathematically lossless.

As long as RED uses REDcode, the data goes through a compression algorithm.


David
 
I think people are ignoring the context of Yedlin's demonstration. He's speaking to professional cinematographers, when can choose between most of these formats (15 perf the exception), with a primary target of standard theatrical/home distribution. His ultimate point is that all formats have the resolving capability of producing sharp and resolute images, so we ought to be more concerned with issues like halation, compression, and scaling than resolution.

But because our favorite camera company didn't "win" (whatever that means in this context), some people will gripe about it.

Steve is basically saying the same thing computer tech guys were saying in the early 00's about the clockrate wars between Intel and AMD. We've got the high clock speeds, we need the rest of the architecture to improve.
 
That is sorta the point I was trying to make earlier. He clearly demonstrates that for this style of photography anyone of the formats has the resolving power for 4K delivery.

He also makes some very interesting observations about perceived resolution, grain, sharpness and viewer experience when considering these factors. I did not think any camera won or lost when looking at the intended frames cross cut with each other. No one in the cinema is stoping the film and having the projectionist punch in to 1:1 resolution.

I think that is also sort of his point. Like when I watched revenant, as a film, as a story, as a piece of art, I could not tell when I was watching Alexa 65 or the mini and I want to still be able to enjoy movies in that way. While I appreciate the geeky pixel peeping as much as anyone, when I settle down to experience a film for the first time, I save the geeky stuff for the second and third time I watch it.

On Netflix 4K or 2k I love the crown in 4K but then again I love Peaky Blinders in HD, go figure, it all works with the story.

David
 
This is a pretty interesting observation that Yedlin makes:

So the idea was, what if I only have 2K of this 6K image — what if I have to choke it all the way down to this. What does that image look like in comparison to the 6K original after it is scaled up to 4K? And then we see how very little the resolution actually matters; it’s almost indistinguishable. And from an audience perspective, it’s no big difference in terms of their perceptual experience. What you can see, however, is that while you get almost no difference from the resolution change, you do get a difference from the scaling algorithms used to go from 6K to 4K and 2K to 4K.

He's made a pretty good case that beyond a certain number, there's not a huge difference between cameras and you can make them all match in post... and that includes film and digital. And I was glad to see he made this observation about "The K's":

If you want your image to look resolute — without scaling or cropping or extracting — you can’t just make a decision based on the Ks. But even if you do want to do an extraction, you still can’t choose just based on the number of Ks.

It would be interesting if Yedlin had been able to test the Panavision DXL and the Varicam LT, both of which I think are game-changers in a lot of ways.
 
I think Steve makes some great points. But as a DP that DOES opt for higher resolution image capture, I think the biggest issue that Steve sheds light onto are all of the OTHER factors affecting the images.

Noise/grain, compression, halation (both camera and lens), scaling, and order of operations. Those don't get talked about enough in my opinion.

When I teach on this topic, I certainly do profess a need for resolution as part of an origination "digital negative," as we can clearly see the Alexa 65 in his tests supporting. Why does the higher photo-site digital camera with a bigger sensor look better than all the rest if resolution doesn't matter?

The higher spatial clarity and lower noise DOES play a huge part in flexibility. That is not disproved in this demo. Rather, what is actually proved is we need to pay WAY MORE attention to the other factors. Noise has a significant impact, compression as well... And as our industry is set up now, cinematographers have less and less control over the post handling of our images. I experience that fact daily.

Watching this demonstration only galvanizes me to be come more diligent in informing productions on the post path, whether I can or cannot be included in that process. The bare minimum I can do is help steer them in the right directions. The best case scenario, I can be intimately involved in that process. However, that's not always possible. Having this video at my disposal will become a tool I will no doubt frequently pass along to help educate others on the importance of the post path our images will be subjected to.

What I will call exception to is the methodology of taking a 4K RGB image, down-scaling it to 2K and then Up-rezzing that image back to 4K to illustrate that we don't need higher resolution cameras. Someone please tell me a full RGB 2K camera in popular use these days! A more appropriate comparison would have been to take a 2K BAYER pattern image and upscale that to UHD and compare that to the 6K Bayer image scaled to UHD. It's a much different beast upscaling an image with full 2k RGB data than one with 1/3 the amount of information.

Steve's test was very interesting and DID show benefits of resolution, as well as dispelled some assumptions on the other factors role in perceived sharpness and acuity. It only supports my current shooting practices in how I hand digital cameras as an origination medium and more importantly, gave me more motivation to try and sway the post processing of my work even more.
 
I think his overall point about resolution is incredibly valid and very well put across. He is talking really about resolution for display. When you are sitting in front of the screen, there is a physical limit as to what you resolve. fact.

The demonstration between contrast vs resolution is a fantastic example of perceived resolution too.

The talk of different aberration is valid. The lens choice will affect the look more than resolution. He's one of the few DPs working that really knows his stuff. I remember a discussion over email with all that display prep stuff from a while back - and he really is totally aware of what he's doing. Anyone that does a presentation in Nuke, enough said :)

It demonstrates how important post it - he talks about multiple scaling operations causing issues and i assume he knows in Nuke that transforms will most often concatenate so there is hopefully only one real transform done at the end of the chain (if the node network is set up that way). I do wonder whether he's on the right Nuke for R3Ds although i suspect he got the R3D from another system.

8K is an awesome format to acquire but should you work in 8K in post? Nope 99% of the time. But what he doesn't touch on is that high res capture when downsampled makes up for bayer issues and improves tonality and also what he doesn't touch on either is movement, or aliasing of hard(ish) edges - so the blinds in the F55 were horribly oversharpened by the camera and that will lead to aliasing issues as the sensor bloom moves from line to line.

But he is looking purely from a DPs view, purely from a film aesthetic and narrative drama being projected in a cinema. 100% valid observations.

cheers
Paul
 
There are a lot of thing in this test that I can clearly see a difference on and I'm always sceptical of tests when using Red since most of the time, even by top professionals, the post processing of R3D files aren't done correctly while compression is often too high.

In this test however, I don't think any of that really matters though. The purpose and subject is to show the perceivable differences for the audiences. While we can nitpick and pixel peep this test, the end result is about what the audience will perceive when watching a story and in that regard Steve Yedlin is right, all the top cameras today have enough resolution.

However, I would have liked to see him go into more detail on how resolution affects other aspects of the image, like color depth and noise (he just scratched the surface on the noise issues). I would have liked him to show the difference on scaling down 8K images to 4K compared to 6K scaled down and 4K native from original sources. Because if we fully follow what he says, then Dragon 6K should be less noisy than Helium 8K since there are larger photosites on the same sensor size on the Dragon 6K compared to the smaller photosites on the Helium 8K. In reality however, the 8K scaled down image gives you less noise. That factor isn't even touched upon.

The bottom line though, is that it's pretty clear that the resolution game needs to stop or slow down. Dragon 6K was enough really, whatever people are pixel peeping about 8K displays and so on, it's only viable for very specific application and in those cases there might just be better to have a special super expensive 12K camera that applies to those gigs, but for regular work we have enough resolution. This is why I didn't upgrade to Helium because I think it wasn't enough progress with the Helium to jump onto it.

The only thing I thought was worth it was the noise, but colors looks pretty much the same as on Dragon. Arri haven't changed their sensor much and their colors looks amazing and has been since the beginning. This is why focusing on color is so crucial and why I think Reds next sensor really needs to tone down the resolution push and focus on actually deliver on the look of the sensor more than before. I like the image of both Dragon and Helium, but compared to the Alexa, the colors feel "flatter", more "smudged" while people's skin really pops on the Alexa. When saying this, people pop out perfectly lit daylight images of people with perfect skin and answer back that skin tones do look beautiful on Dragon and Helium but that is such isolated cases of shooting that it doesn't apply to the real world use. Alexa looks good in almost all conditions while Red cracks down whenever the light isn't close to perfectly lit. The differences are marginal of course, but it's still enough for me to feel underwhelmed every time I'm color grading my Red shots compared to when I get gigs shot on Alexa.

I'm not sure what Red should do next with their sensors, but since most cameras now aims to become full frame, that puts it in somewhat of a problematic position. Dragon 8K I get, that sensor was a real bomb and it's in my opinion the best sensor Red has produced, far beyond Helium. But since Helium is already 8K, that makes it problematic to carry over to full frame. I would rather see Red scale back on resolution per mm of the sensors to deliver a sensor with focus on improving colors while having full frame 8K. It feels like they've given up on the Vista Vision part in favor of Helium. I'd rather see something FF 8K that rivals what we see Arri doing with Alexa 65.
 
Christopher Probst said:
I.. And as our industry is set up now, cinematographers have less and less control over the post handling of our images. I experience that fact daily.

.
Click to expand...

Which is pretty bizarre, when you think about it.
Imagine somebody would have told Ansel Adams, that he is not allowed (or qualified) to process his own shots in his lab.

IMHO post production is an important part of the image making process, and can even make or break your material.
I think every DP should have a deep understanding of post production, if not the ability to do it him self.

Personally, I really hate it, when I can't do the color correction of my material.
It usually comes out totally different, than I had in mind, while shooting.
 
Frank Glencairn said:
Which is pretty bizarre, when you think about it.
Imagine somebody would have told Ansel Adams, that he is not allowed (or qualified) to process his own shots in his lab.

IMHO post production is an important part of the image making process, and can even make or break your material.
I think every DP should have a deep understanding of post production, if not the ability to do it him self.

Personally, I really hate it, when I can't do the color correction of my material.
It usually comes out totally different, than I had in mind, while shooting.
Click to expand...

yup. And the big concern is budget. I guess that has to get negotiated as part of the original contract to add days for post production and grade approval. There are many ways to attempt to achieve the intended look, with references, notes, etc, but you have to have a production and post depeartment wanting to interact.

David
 
David Battistella said:
yup. And the big concern is budget. I guess that has to get negotiated as part of the original contract to add days for post production and grade approval. There are many ways to attempt to achieve the intended look, with references, notes, etc, but you have to have a production and post depeartment wanting to interact.

David
Click to expand...


I can just say that there are also other factors in play. A lot of directors, producers and agency producers dont always want the dps input in post, just as they usually dont want actors stepping into the editing room and starting to tell the editor what take of their acting should be used.

Harsh I can agree, but that is a very comon scenario even more so in comnercials. So dps that write in their contracts that they need to be involved in the post process are possibly wellcomed by some but not by all.
Sure there are productions that wellcome the dps input during post but in the end its usually not the DP that has the final word on grade, VFX or any of the post work.
 
I always find fascinating the different approach to post when working with commercial productions in US and in Europe. Here in US, I rarely, if ever see or talk to a DP. Post is pretty much an ad agency domain. On the other hand, just about every time I do a job in Europe, it is not only run by a director, but it is also almost always attended by a DP. At the very least, if the DP or director must run to the next gig, we usually get to do at least a couple of hours "look" session, which becomes the basis for the final grade. The whole process working with director and DP makes so much more sense and the result is always much more streamlined. On the other hand, in US the post is kinda strange, as agency creatives make things up as they go along, pretty much negating a lot of good work done by DP during the production. So, I hope US production does get involved in post a bit more, if they would like to have more say in the final outcome, even if it's not always possible. Although, as I mentioned, European productions somehow always find a way for that to happen...
 
jake blackstone said:
I always find fascinating the different approach to post when working with commercial productions in US and in Europe. Here in US, I rarely, if ever see or talk to a DP. Post is pretty much an ad agency domain. On the other hand, just about every time I do a job in Europe, it is not only run by a director, but it is also almost always attended by a DP. At the very least, if the DP or director must run to the next gig, we usually get to do at least a couple of hours "look" session, which becomes the basis for the final grade. The whole process working with director and DP makes so much more sense and the result is always much more streamlined. On the other hand, in US the post is kinda strange, as agency creatives make things up as they go along, pretty much negating a lot of good work done by DP during the production. So, I hope US production does get involved in post a bit more, if they would like to have more say in the final outcome, even if it's not always possible. Although, as I mentioned, European productions somehow always find a way for that to happen...
Click to expand...

In Europe, everything is about cooperation. In the US, everything is about plausible deniability ;-)
 
It' in the (European?)director of photography job description to be responsible of the picture quality up to the delivery.

Pat
 
David Battistella said:
Just as an FYI, REDcode uses wavelet based compression. 1:1 is not possible with REDcode. If you want to see "uncompressed" that is to say mathematically uncompressed images from a RED sensor, then shoot some frames using REDcode burst at 2:1 compression. Those shots would be mathematically lossless.

In other words, if it goes through REDcode it is always compressed, but at 2:1 the compression The data is mathematically lossless.

As long as RED uses REDcode, the data goes through a compression algorithm.


David
Click to expand...

I know and I would like to have a choice to go 1:1 RAW and/or 1:2..1:?? .R3D when DSMC3 comes out.
 
Misha Engel said:
I know and I would like to have a choice to go 1:1 RAW and/or 1:2..1:?? .R3D when DSMC3 comes out.
Click to expand...

...And all I am saying that 1:1 REDcode is impossible. As in, not possible. As in, it can never happen. Just informing you that as long as there is REDcode you will not have a 1:1 choice/option.

David
 
You must log in or register to reply here.
Back
Top