Is the Alexa better than the F35?

Steve the F35 is built from the ground-up to do only 1080P, the way the sensor is laid out, the way the processors work, etc... is all setup for 1080P.

With a Bayer pattern sensor you have more flexibility about what resolution you choose to deliver, obviously ARRI are choosing to deliver somewhat higher than their sensor resolution. This is not uncommon in consumer cameras, but very unusual for a professional camera and extremely unusual for ARRI - the company that took a long time to call the ALEXA anything other than 2K.
 
steve phillipps said:
I thought the F35 had more pixels / greater resolution than the Alexa / Amira? Does that not mean that if the Amira can up=rez to 4k acceptably the F35 can too? What am I missing?
Steve
Click to expand...

It's a little bit more complicated. The F35 isn't wired in a way that allows you to collect photosite date - rather, it processes the photosite data and provides a 1080p signal. Vertically, there are columns of 2160 photosites, but every pair is averaged together (probably with some extra math for edges) by the DSP to get the 1080p signal you see.

Without replacing the DSP you're not going to get access to this photosite data on the F35.

The F35 will probably do quite well with a simple 4k uprez. But it would benefit even further from a specialty algorithm that takes into consideration the RGB stripe sensor.
 
steve phillipps said:
Thanks James - but why does the Bayer pattern sensor have more flexibility?
Steve
Click to expand...

It's worth spending some time reading up on sensor types. The F35 uses an RGB stripe sensor, while most single chip digital cinema cameras use a bayer pattern. "3 chip" cameras don't typically use a pattern - a prism sends each sensor a specific color of light to process.

3 Chip cameras with prisms have to account for fringing and chomatic abberation - that is, certain wavelengths of light bend at different rates - but essentially, you're going to be able to see each color of light at each pixel position.

However, single chip cameras have a fundamental problem - you can't collect more than one color at a specific spot (there are a couple patents that you could argue about this with, but those aren't relevant to the cameras we currently have).

So, you have to have some sort of pattern. The most common one is the Bayer pattern. It's in the Alexa, Amira, Sony F3, RED, etc. It tends to have a layout something like this:

G R G R G R G R G
B G B G B G B G B
G R G R G R G R G
B G B G B G B G B

You'll notice there are twice as many green photosites as red or blue. When manufacturers of cameras talk about how many photosites their raw camera sensor has - this is what that raw looks like. The problem is, it's not a usable image yet. It needs to be debayered - interpreted if you will. And their are numerous algorithms available to do it with. It's the reason why RED users talk about saving raw footage and going back and debayering at a later date. Particularly in the early days of the RED One the debayering algorithms were not mature, but if you held onto the raw footage, you could come back later and see what a better algorithm might come up with.

One of the simplest ways to interpret bayer data is to break it into a four photosite block:
G R
B G

I have a red value, a blue value, and two green values. I add the green values together, then divide by two. I now have a value for each color, and I could use this as a video image pixel. However, the results are fundamentally guesses (or perhaps assumptions). I only collected blue from one of four quadrants. What if at the other three photosite locations the values of blue would have been very disparate from the one quadrant I collected from? In that case my sample wouldn't be representative.

A solution is to compare my blue sample against the blue samples received nearby. If the nearby samples are similar, it's likely safe to consider the sample representative. If not, there may be an edge or fine pattern.

It gets very complicated.

By contract, the F35 RGB stripe sensor look like this:

R G B R G B R G B R G B
R G B R G B R G B R G B
R G B R G B R G B R G B
R G B R G B R G B R G B
R G B R G B R G B R G B
R G B R G B R G B R G B
R G B R G B R G B R G B

And a single pixel of 1080p comes from six photosites on the F35:

R G B
R G B

(Please note, the photosites aren't square)

As you can see, this pattern gives us an equal sampling of red, green, and blue. You still run into the problem of not having a sample of every color at every location, but the double sampling gives some protection against outlier results, and gives a degree of internal edge detection vertically. Further, I'm confident the camera is capable of doing nearby photosite comparisons for further edge detection.

The RGB stripe sensor shown here provides excellent 1080p results, better than any dedicated 1080p bayer sensor. But it's not really intended for upresolution beyond 1080p.
 
By the way, the Sony F65 is an odd duck because while it does have a Bayer pattern, it's arranged diagonally, and interpreted in a different manner.


<G>:<G>:<G>:<G>:<G>:<G>:<G>:<G>:
~:<R>:<B>:<R>:<B>:<R>:<B>:<R>:
<G>:<G>:<G>:<G>:<G>:<G>:<G>:<G>:
~:<R>:<B>:<R>:<B>:<R>:<B>:<R>:
<G>:<G>:<G>:<G>:<G>:<G>:<G>:<G>:
~:<R>:<B>:<R>:<B>:<R>:<B>:<R>:
 
Sony's counting of the F65's sensor is also odd, in reality the sensor is more like 6000 x 3000 but Sony count it as 8000 x 2000, presumably for marketing reasons.
 
damn awesome thread guys. awesome. I love the scrapbook.
I love the heated opinions.

I am also a CCD fan over CMOS. Motion feels better. How it sees motion - and grain - its natural grain. I love grain so much. And that's why I like the Red One MX over the Red Epic and prefer to shoot with the Red One MX on most of my shoots that have the budget for the epic. Just like my man Fincher liked the grain of the Red One over the Epic.

With that said, Elysium is the best looking Red film I've seen. I also loved the look of Winter's Bone and District 9 and that post apocalystic movie with Denzel Washington. And the Canyons I can easily say is the worst looking Alexa movie I have ever seen.

And I've seen some movies shot on Kodak and Fuji that look so so bad.

So boo hoo to that.

Also agreed - lets not turn this forum into a red user style fanboy bashing forum - everyone has their opinions and they are all valid. Not even can we quantify dynamic range - your usuable dynamic range is different than mine. So we are all right. We all are winners!!
 
I love Elysium too, looked fantastic.

I have to say I do enjoy it when people give me reference films from before approx 2000. I find most movies since then have been graded so heavily, or have so much CGI work, that getting anything near that look on set is going to be impossible. Older films had fewer tools available, all the look came from filters, in-camera FX, lighting and set design - how it should be (IMHO). I love when people reference those because I have half a chance of getting that look on set!
 
Steve,

It isn't so much that you can or can't, but the RGB stripe (a unique design for the F35/Genesis) is designed very specifically for 1080P output.

As you can see from the wonderful illustrations above, the bayer pattern is designed to try and simulate a fairly random pattern without actually being a random pattern. People and company opinions vary on what the actual "recoverable" resolution from a bayer sensor is. This is why RED courted a bit of controversy when they came out. In short, or as short as I can be:

The term "4K" derives from the film scanning world, where a film scan was performed to a resolution of 4096 pixels wide.

There was a green, red and blue pass on that scan. So, 4096 green 4096 blue and 4096 red.

In the display world, a single pixel has three sub-pixels, red green and blue. So scanning film like the above gives you all the info you need for genuine "4K" acquisition.

With the RED, or now many other "4K" cameras, their sensors had 4096 photosites across. However, because photosites work in black and white, you need a colour filter array to help derive colour.

In the bayer pattern world, half of the pixels will be green, the other half red or blue depending on the line you're on.

There is very clever maths to get a full colour picture out of the bayer pattern and it works quite well. However, if you were a numbers man, you'd look at it like this:

Film's "4K":

4096 Green
4096 Red
4096 Blue

Total = 12,000 photosites

Red's "4K"

2048 Green
1024 Red
1024 Blue

Total = 4,000 photosites

As you can see, there's a huge difference. Now, the difference never felt as big for many reasons. A cynical person might say the RED or similar is capable only of a 1K image, because that is the lowest common denominator. Actually, it works much better. Partly because we are much more sensitive to green than red or blue, partly thanks to amazing debayering algorithms (part of the reason it takes so long to process raw footage). In reality, the Bayer pattern can generate anywhere from 50 to 90% of the claimed resolution, depending on who you ask. When you "de-bayer" it, you have flexibility as to what your end resolution is going to be. You can even make it larger than the original sensor, although this is an odd choice to make.

I personally always felt the RED (one) made nice 2K pictures but was not capable of making best use of 4K. As 3K is not an established format, I'll ignore that. Others may, and will, disagree with me. My "proof" was always comparing RED's 2K with an established 1080P 3-chip camera of the day. The comparison was night and day.

Now, Panavision and Sony decided at the time that 1080P was good enough for cinema if you used everything the format was capable of. They decided to give each pixel in the finished picture not one, but two each of red, green and blue. There are two to satisfy the Nyquist theorem. Doing this gave them stellar 1080P pictures but it means the camera is locked into producing that one format.

I suggest for further research, find Panavision's seminar "Digital camera specifications demystified" - it is very enlightening.
 
Thanks James, I seem to remember that I have seen that Panavision seminar. Still a bit unclear though. I suppose all I'm trying to establish, ignoring all numbers and theories, is how well the F35 image would slot into a film shot on say F55 or Epic at 4k and delivered / viewed at 4k.

Steve
 
Steve

In terms of F55, it would do it very well.

Epic shot at 4K it would also blend quite well.

When you look at the raw numbers the F35's sensor is higher res than the F5/55 and very similar to the EPIC M-X, certainly higher res than the RED M-X (which couldnt use the full sensor).

James
 
That's the bit I don't get though - you say that the F35 has higher res than F55, but previously that it wouldn't up-res well due to the RGB stripe thing?
Steve
 
okay, I'll try and explain it best I can.

The F35, despite having a higher resolution sensor than the F55, is designed to go to 1080P, no matter what you do. That is what the RGB stripe effectively does, it locks you in to one set resolution.

Now, the F55, having a "4K" sensor in a bayer pattern, is able to derive a higher measured resolution than the F35 because of the way it has been setup.

Think of it this way - the F55 is able to give you a "70%" of 4K image, but the F35 will give you a "100%" of a 1080P one. The reality is that 1080P/2K is more than adequate for the vast majority of productions, but most people don't realise that.

The F35's 1080P is so good that were you to up-res it to UHD, it would look very good. But, it will always have to go through that 1080P output, which means you'll never be able to get better than 1080P.

I hope that makes some sort of sense?

We sadly live in a fickle world where most end-users/producers/etc... have become obsessed with one number and don't understand anything else. Everyone is going nuts for 4K acquisition on TV, but last time I checked "HD" TV was 1280x720p (or1080i) 4:2:0 at a very low bit rate - not even showing off the best of what HD can do.
 
JMtheDP said:
okay, I'll try and explain it best I can.

The F35, despite having a higher resolution sensor than the F55, is designed to go to 1080P, no matter what you do. T
Click to expand...

But sony was speaking of 2K upgrade for F35.:happy:
 
steve phillipps said:
That's the bit I don't get though - you say that the F35 has higher res than F55, but previously that it wouldn't up-res well due to the RGB stripe thing?
Steve
Click to expand...

Steve I think you are confusing and James is maybe confusing you with long explanation. James is saying you can't up-res in camera with F35 because of 3 strip sensor. But nothing stop you up-res in post and I think would look very good.
 
SJX said:
Steve I think you are confusing and James is maybe confusing you with long explanation. James is saying you can't up-res in camera with F35 because of 3 strip sensor. But nothing stop you up-res in post and I think would look very good.
Click to expand...

No, I do get that, just wondering how good it would look at 4k delivery especially when intercut with a "proper" 4k camera. Guess the only real way would be to do it and have a blind test.
Steve
 
Steve

That's kinda the point I'm trying to make. If you were cutting the F35 with a camera that did 4K as well as the F35 does, you'd notice a difference (but only viewing at 4K!).

However, you can certainly get away with cutting it with the current crop of cameras. A little grade can go a long way. I doubt you'd be mixing within the same scene, but within a film I'd have no issues.

Also @ SJX, I don't think Sony will be coming out with any more upgrades for the F35. I don't know how they planned to get 2K out of it, but 1080P/2K is not really a big difference anyway. In fact 1080P will scale to UHD (which is what most people mean when talking about 4K) better anyway.
 
You must log in or register to reply here.
Back
Top