4k ana mode : disappointing resolution

[Karim_Nassar]

Hello everyone,

So I tested the 4k ana mode on my mx.
Very nice to have desqueezed footage straight out of the camera.
The workflow is great.

But when I watched the footage on my computer I was disappointed at how bad it looked compared to my 4kws.

So I looked into it and pulled this comparison:

No wonder the 4k ws footage looked better!

On the epic, what is the best mode for anamorphics and how much of a resulting resolution would you get?

thanks!

 

[Sulekh Suman]

The drop in resolution is being addressed to a some extent in Dragon(6K Aanamorphic mode). A 4/3 sensor will allow for maximum resolution but that will be another sensor another year..

 

[Karim_Nassar]

The drop in resolution is being addressed to a some extent in Dragon(6K Aanamorphic mode). A 4/3 sensor will allow for maximum resolution but that will be another sensor another year..

If I am reading the dragon diagram correctly the height of the 6k ana mode will be 3160px.
Do we know the width yet?

 

[Ignacio Aguilar]

Bear in mind that spherical lenses tend to perform better than most anamorphics out there, specially compared to vintage glass such as Lomos, Kowas and/or converted Zeiss Standard/Superspeed, Cooke S2/S3, etc.

When used with 4-perf 35mm in the old days, anamorphic used a larger area of the negative for 2.40:1 than the Super 35 format and this not only hide the not-so-good performance of the lenses, but also could result in a sharper projection in the case of Panavision "E", "G" and Primo Series, Hawk "V Series" or other modern lenses. But these days, with 16x9 sensors, anamorphic is not about resolution anymore, it's about its special look, aberrations, distortions, flares and a softer look due to the optical characteristics of the lenses and the lower resolution of the sensors in anamorphic mode.

 

[Peter Majtan]

Is ir just me are are you doing it the other way round? Instead of "desqueezing" you are actually "squeezing" the footage even further...
Instead of taking the 2816 x 2304 and squeezing it to 2816 x 1152, try desqueezing it to 5530 x 2304. Then apply sharpening, your grade, etc... - and finally scale down to 4096 x 1706 (inside 4096 x 2160 letterbox) for 4K DCI master. You will end up with far superior footage...

On another note - I have been hearing a lot lately about new breed of 1.2 and/or 1.33 anamorphic glass for the new generations of 2:1 and 16:9 sensors. I personally could not care less - I am not a big fan of anamorphic lenses...

:sifone: Peter

 

[Karim_Nassar]

Bear in mind that spherical lenses tend to perform better than most anamorphics out there, specially compared to vintage glass such as Lomos, Kowas and/or converted Zeiss Standard/Superspeed, Cooke S2/S3, etc.

When used with 4-perf 35mm in the old days, anamorphic used a larger area of the negative for 2.40:1 than the Super 35 format and this not only hide the not-so-good performance of the lenses, but also could result in a sharper projection in the case of Panavision "E", "G" and Primo Series, Hawk "V Series" or other modern lenses. But these days, with 16x9 sensors, anamorphic is not about resolution anymore, it's about its special look, aberrations, distortions, flares and a softer look due to the optical characteristics of the lenses and the lower resolution of the sensors in anamorphic mode.

I did read about that, there is a good illustration of the total use of film with anamorphics on wikipedia.
People usually define anamorphic by the flares and bokehs like you said but to me it is all about the wide angle. I really like the wide field of view.

Is ir just me are are you doing it the other way round? Instead of "desqueezing" you are actually "squeezing" the footage even further...
Instead of taking the 2816 x 2304 and squeezing it to 2816 x 1152, try desqueezing it to 5530 x 2304. Then apply sharpening, your grade, etc... - and finally scale down to 4096 x 1706 (inside 4096 x 2160 letterbox) for 4K DCI master. You will end up with far superior footage...

On another note - I have been hearing a lot lately about new breed of 1.2 and/or 1.33 anamorphic glass for the new generations of 2:1 and 16:9 sensors. I personally could not care less - I am not a big fan of anamorphic lenses...

I know of this technique that I have used many times.

But I have no idea why people are recommending it.

There is no detail gain what so ever, the image is simply bigger.

Now you do get a little bit of sharpness when scaling back down, but this is artificial sharpness from the interpolation that does not reveal more actual detail.

Why would you get 5632 pixels level of detail from 2816 actual pixels of the sensor?
Even if the image is squeezed by the anamorphic lens, in the end one pixel of the sensor will record one color.
It will not record 2 colors because of the 2x squeeze of the lens and then reveal that pixel information once you desqueeze.

At least that is my understanding of it, and from my personal footage and tests I have not seen any benefits from doing this aside from a bigger yet not more detailed file.

Now I am a beginner and I would be happy to be proven wrong and hear more on the subject

 

[Freya]

You appear to be reducing the height, thus losing vertical resolution from the original, instead of expanding the width.

I'm not sure why you would want to rescale twice like the other poster proposes tho, that seems like an extra unnecessary scaling?

Freya

 

[Andreas Mendritzki]

I did read about that, there is a good illustration of the total use of film with anamorphics on wikipedia.
People usually define anamorphic by the flares and bokehs like you said but to me it is all about the wide angle. I really like the wide field of view.



I know of this technique that I have used many times.

But I have no idea why people are recommending it.

There is no detail gain what so ever, the image is simply bigger.

Now you do get a little bit of sharpness when scaling back down, but this is artificial sharpness from the interpolation that does not reveal more actual detail.

Why would you get 5632 pixels level of detail from 2816 actual pixels of the sensor?
Even if the image is squeezed by the anamorphic lens, in the end one pixel of the sensor will record one color.
It will not record 2 colors because of the 2x squeeze of the lens and then reveal that pixel information once you desqueeze.

At least that is my understanding of it, and from my personal footage and tests I have not seen any benefits from doing this aside from a bigger yet not more detailed file.

Now I am a beginner and I would be happy to be proven wrong and hear more on the subject

But this is the whole point of shooting Anamorphic (always has been...) you use the squeeze of the lens to fit more 'information' onto fewer pixels/negative area. So while yes, you're dealing with less actual pixels, the 'effective' resolution, and the proper way to de-squeeze it, is to 5632 x 2304. After all, its a 2x squeeze horizontal squeeze lens, not a 0.5 vertical squeeze lens we're talking about!
So while I'd be willing to bet that a cropped 5632x2304 spherical image would appear 'sharper' than a de-squeezed 2816x2304 image, the sharpness from your 4k ana mode shouldn't be all that less than from your 4.5k ws crop.

What lenses are you using to do your comparative tests?

 

[Karim_Nassar]

You appear to be reducing the height, thus losing vertical resolution from the original, instead of expanding the width.

I'm not sure why you would want to rescale twice like the other poster proposes tho, that seems like an extra unnecessary scaling?

Freya

I am losing vertical resolution, that is correct.

But the issue I have is that I don't see how we are gaining resolution by stretching the pixels horizontally.
To me (and I might be wrong) we are simply stretching which results in a bigger file but not a higher res file (imo).

But this is the whole point of shooting Anamorphic (always has been...) you use the squeeze of the lens to fit more 'information' onto fewer pixels/negative area. So while yes, you're dealing with less actual pixels, the 'effective' resolution, and the proper way to de-squeeze it, is to 5632 x 2304. After all, its a 2x squeeze horizontal squeeze lens, not a 0.5 vertical squeeze lens we're talking about!
So while I'd be willing to bet that a cropped 5632x2304 spherical image would appear 'sharper' than a de-squeezed 2816x2304 image, the sharpness from your 4k ana mode shouldn't be all that less than from your 4.5k ws crop.

What lenses are you using to do your comparative tests?

This was the point and made complete sense on film:

because you are actually physically gaining resolution on the top of the film.

If my understanding is correct this is what is happening on a digital camera sensor:

the image is squeezed in the same sensor pixel area so I don't understand how we would gain resolution by stretching.

Am I wrong here?

I need to clear this.
So I will shoot a video in my studio with the sharpest image possible and I will post the results here :wink5:

 

[Freya]

Well you won't be gaining resolution by using anamorphic lenses on the red camera but by shrinking it, like you appear to be doing, you are losing resolution, not on the horizontal plane but on the vertical (where there is also colour information and all the rest!)

Something seems to have gone wrong with your lower two images, just above BTW.

Freya

 

[Freya]

You need to unsqueeze the video not shrink it. If you shrink the video you are losing the vertical information. You won't gain resolution from the anamorphic lens but you won't lose any like you do if you shrink it down.

 

[Karim_Nassar]

You need to unsqueeze the video not shrink it. If you shrink the video you are losing the vertical information. You won't gain resolution from the anamorphic lens but you won't lose any like you do if you shrink it down.

Regarding the supposed resolution gain that we seem to agree on:

When you shoot 4k ana on the mx, you shoot at 2816/2304 effective pixels.

If we follow the reasoning that the 2x ana lens actually gives you more detail and you should desqueeze the width; you get a 5632/2304 pixels image.

Which means that by doing that, you are actually getting more resolution than when shooting in 4k16:9 mode which is 4096/2304 pixels.

I don't see how that is possible. (if I'm wrong tell me)

Regarding the loss of resolution vertically by squeezing the footage vertically:

Obviously we lose resolution vertically, but we are stretching horizontally if we don't.
So what I am wondering is what good is this doing to the image quality?

Would we not get better image quality by directly squeezing vertically?
Instead of stretching horizontally and then shrink the image back down anyways?

Need to test again but I have yet personally not seen a benefit to doing that.

 

[Freya]

If my understanding is correct this is what is happening on a digital camera sensor:

the image is squeezed in the same sensor pixel area so I don't understand how we would gain resolution by stretching.

Am I wrong here?

I need to clear this.
So I will shoot a video in my studio with the sharpest image possible and I will post the results here :wink5:

Only just worked out what you meant. I thought you made a mistake with the images. The anamorphic modes on the red are a different aspect ratio to the normal modes, so as to give the correct aspect ratio when unsqueezed The image is not squeezed onto the same pixel area but a smaller one to if you were shooting flat. So less resolution.

The process you are following tho shrinks the video even further losing further resolution again.

It's a shame there is no 4.5k anamorphic mode on the RED One, it should be possible.

Freya

 

[Karim_Nassar]

It's a shame there is no 4.5k anamorphic mode on the RED One, it should be possible.
Freya

The mx has 4'520 by 2540 active pixels.

Currently we have a 2816 by 2304 ana mode.

So by keeping the correct aspect ratio in theory we could get a maximum of a 3104 by 2540 image.

 

[Freya]

Regarding the supposed resolution gain that we seem to agree on:

I think I'm agreeing that there is a loss of resolution as compared to shooting flat, not a gain!

Would we not get better image quality by directly squeezing vertically?
Instead of stretching horizontally and then shrink the image back down anyways?

Now you have confused me!
I only get those two options? (Maybe you should stand for government) neither of those seem right to me!

I guess the big question you have to ask is what resolution you are finishing at?

It's especially confusing because that video will be a 16:9 size presumably and the video will have to be letterboxed into the 16:9 frame.

The example of 4096 x 2160 for a 4k DCI (not sure if that is the same as a DCP?) was given before, with 4096 x 1706 inside that, so if you start with 2816x2304 then you need to unsquish it to 4096 x 1706. You are throwing away some info still but not as much as if you went down to 1152.

It's definitely better to scale up and then down than down and then up. You appear to be going for a scale down and then back up again. I suspect it's even better still if you can go direct to the finished resolution with as little scaling as possible but maybe it doesn't work out that way!

Freya

 

[Peter Majtan]

Karim - it is not about gaining horizontal resolution - it is more about not loosing vertical.
Doing what I have recommended in my first post will yield picture with far more real resolution then the way you do it.
Of course it won't be "true" 4K, but much better then your version.

Just please try this on one frame and compare the two results...

:sifone: Peter

 

[Karim_Nassar]

I think I'm agreeing that there is a loss of resolution as compared to shooting flat, not a gain!



Now you have confused me!
I only get those two options? (Maybe you should stand for government) neither of those seem right to me!

I guess the big question you have to ask is what resolution you are finishing at?

It's especially confusing because that video will be a 16:9 size presumably and the video will have to be letterboxed into the 16:9 frame.

The example of 4096 x 2160 for a 4k DCI (not sure if that is the same as a DCP?) was given before, with 4096 x 1706 inside that, so if you start with 2816x2304 then you need to unsquish it to 4096 x 1706. You are throwing away some info still but not as much as if you went down to 1152.

It's definitely better to scale up and then down than down and then up. You appear to be going for a scale down and then back up again. I suspect it's even better still if you can go direct to the finished resolution with as little scaling as possible but maybe it doesn't work out that way!

Freya

I agree with what you said, if you are going for a 4k or a bigger than the original size file you should definitively desqueeze horizontally.
No argument there.


Bottom line of what I am saying is this:

Yes you can desqueeze to get a 5632/2304 image.
But the actual detail you have is what it was shot on 2816/2304.

At least that is what I have noticed and I am off to some testing to confirm.

 

[Karim_Nassar]

Karim - it is not about gaining horizontal resolution - it is more about not loosing vertical.
Doing what I have recommended in my first post will yield picture with far more real resolution then the way you do it.
Of course it won't be "true" 4K, but much better then your version.

Just please try this on one frame and compare the two results...

:sifone: Peter

I agree on the vertical resolution losing part.
And yes off to some testing :001_smile:

 

[Douglas Underdahl]

My understanding is this:

The scope format was invented to make the screens in movie theaters wider, as a way of getting people away from their TVs.

One can make the image "wider" simply by cropping the top and bottom of the image and enlarging the resulting frame: this is called 1.85 or flat, in projection lingo.

But if you use a scope or anamorphic lens, you can use all the negative, even that which was lost when the soundtrack was introduced. This area is a large squareish area that is then stretched 2x by the projection lens. Added benefit is that the projector gate is much larger and the image on the screen is much brighter. More negative (allmost all of it, in fact) means less grain. Special splicers have to be used for the negative.

Now folks want those scope artifacts. To get those, they get scope lenses (I have a nice 35mm Lomo) and put it on their digital camera. But these sensors are smaller and are not squareish, but rather 16:9. So the sides are not used. And, some of the top isn't used either because the height of the Red One and Epic/Scarlet sensor isn't as tall as the height of the 35mm flim scope area negative.

So when you do a desqueeze in post, aka a stretch, usually 2x, you are using a cropped/smaller area of the sensor, and this gives you less resolution. Also as was said before, scope lenses just aren't as sharp as spherical to begin with.

A bit of sharpening in post helps. I think with the Red One, all scope material is about 2.5k. With the Epic and it's 5k ana mode, perhaps it's more like 3.5k?

 

[Andreas Mendritzki]

I agree on the vertical resolution losing part.
And yes off to some testing :001_smile:

Exactly:

2.4:1 SPHERICAL Cropped from a 16:9 sensor = resolution 'loss' on the top and bottom
2.4:1 Anamorphic image from a 16:9 sensor = resolution 'loss' on the left and right.

Example (using aprox numbers by the way, but the math should be correct):

5k: 5120 x 2700 = 14MMP

SPH Crop to 2.4:1 ratio from 5k:
5120x2132 = 10.9MP

Ana Crop:
3240 x 2700 = 8.7 MP
De-Squeezed:
6480 x 2700 = 17.5 MP "Effective Resolution"

So what you get is more "actual" resolution by cropping top and bottom on a spherical lens. However, you theoretically get a bigger image with more "effective" resolution vis-a-vis the anamorphic de-squeeze.

Again. Shoot some tests and see what you prefer.

At this point, shooting ana from 16:9 sensors vs. 4:3 has more to do with field-of-view than resolution if you ask me.

A