Canon vs. RED

Status
Not open for further replies.
ChipG said:
I'm not even sure if number 3 from the OP's list is even necessary, red's 'raw' is actually 4:2:0 and very highly compressed but for some reason they like to call it raw.
Click to expand...

mcgeedigital said:
Chip, I don't know where you get your info, but RED's RAW is NOT 4:2:0.
Click to expand...

Well, here we're confusing the recording format, with what's getting recorded *in* that format.

Red's sensor is limited by its Bayer technology to producing images that can't ever exceed 4:2:0, at its nominal resolution. With a 4K Bayer chip, you *cannot* get 4K with anything higher than 4:2:0 out of it. So no recorded RedRaw image is going to have more than 4:2:0 color in it. So by that metric, Chip was right in the first place.

However, it of course is more confusing than that, as Shao points out very eloquently:

shaocaholica said:
RAW doesn't have the same amount of data for all 3 channels due to the bayer pattern. You'd be wasting bits using a "444" pattern. You really can't talk about RAW in terms of color subsampling because there is none. If its RAW, you've sampled all the pixels even if there are more green ones. Think of it this way, if the sensor is natively recording in a 422 pattern, why waste the bits by oversampling into a 444 stream?

For all intents and purposes, RAW data, before de-bayering, is actually a monochrome image. Color subsampling a monochrome image doesn't even make any sense.
Click to expand...

Redraw does store the raw data. It compresses it, so it's not truly "raw" anymore, but it's a compressed version of raw data, which is indeed a monochrome image. It needs to be de-bayered in order to become a color image.

And during that debayer process, the resolved luminance information does decrease (in Red's case, by a factor of about 20%). So you get about 3.2k worth of detail. But the color res doesn't decrease, so technically you could argue that the resulting image is slightly higher than 4:2:0 color sampling, but nowhere near 4:2:2. I think Graeme put it as 3.2:2:0 at one point, or something similar to that.

The differentiating factor here is like Shao said -- the codec isn't doing any color subsampling at all, it's just storing the raw data (in a compressed format). If it was a three-chip system, that codec could indeed be storing 4:4:4 worth of data. But the nature of the single-chip system is what's limiting it to no higher than 4:2:0 of actual image data.
 
grimepoch said:
I'd say the OP is also missing another important fix that needs to happen, and that's faster read/reset. The MX sensor has a read reset of 5ms, and film has an equivalent of 4ms.
Click to expand...
But film doesn't scan the way a CMOS sensor does, so the comparison is largely irrelevant. Film's shutter works more like a global shutter than a CMOS rolling shutter, because the film is exposed globally for much of the time, and only has a little bit of rolling at the beginning and end of the exposure cycle.

Saying "MX is 5ms and film is 4ms" is misleading at best.
 
Barry_Green said:
But film doesn't scan the way a CMOS sensor does, so the comparison is largely irrelevant. Film's shutter works more like a global shutter than a CMOS rolling shutter, because the film is exposed globally for much of the time, and only has a little bit of rolling at the beginning and end of the exposure cycle.

Saying "MX is 5ms and film is 4ms" is misleading at best.
Click to expand...

That may have been true in the past, but that's my whole point. You are describing the problem that has occurred due to slow CMOS designs. The only reason that CMOS has to do what is has been doing is the read/reset is so long, you spend more time not being globally exposed so you can get the data off the censor.

Let's say you are filming with a shutter open for 1/48, which is 20.8mS With a 5mS read/reset, 15.8mS of your exposure time you are globally open.

That's a SIGNIFICANT improvement over the CMOS censors on DSLR.

The same thing happens with film, and especially fast film. If it takes 4mS for the shutter to go black, in that 4ms with speed rated at ISO320, you are going to get the same problem you would get with a CMOS sensor that was rated at ISO320 if it could have a 4mS read/reset.

Even further, it depends on the CMOS design. Most CMOS designs that are high performance read out a whole row at once, including the red designs. the rolling shutter is between lines reading speed and is exactly equivalent to what the shutter is doing in a motion picture camera.

The point is, 5ms is the time it takes to read the whole sensor. This is a huge improvement over all the DSLRs that are currently sold today. This is clearly one of the worst problems with DSLR as you have to be so careful with the camera not to get jello for non-locked down shots.
 
Has anybody ever taken a Red with its 5ms sensor, and a film camera, side by side and tried to get jello out of them both? Or put long lenses on them and done a skew test?

(and no, I don't care about the famous picture of a car skewing; that's a STILL photo with a sliding focal-plane curtain shutter, that has no bearing at all on what we're discussing here).

I don't care about the math and the numbers, I want to see if the red's image behaves the same as the film image does. If it actually does, and I strongly and severely doubt that it does, then ... okay. But I don't believe a rolling shutter mimics the behavior of a rotating mirror shutter.

Hey Kholi, dig up a film shooter out there and go do this side-by-side. Put them on exactly the same field of view, ideally mounted on the same tripod, on super-telephoto, and skew 'em around. Let's see what the resulting frames look like...
 
You're going to have to wait on the Epic and the other Scarlets to use the faster shutter. The RED ONE upgraded to MX does not have the 5ms read/reset. I cannot tell you why, or what limitations prevent it, only what I've been told.

What we do know is that the RED ONE I believe has a read/reset of 8ms and the Epic 5ms as reported by RED. Until anyone can get their hands on these cameras, we won't know for sure. But I'd say that is a pretty significant speed increase given the read/reset is almost half the time.

And I absolutely agree that these tests are going to be important. But I disagree with you that the rolling shutter doesn't mimic the rotating mirror shutter to a large degree. You've got a spinning disc that has an edge that moves across the face of the film at the start and end of exposure. I'll give you that it is not exactly the same since that edge actually rotates slightly as it moves across the film, but if it takes a certain amount of time for that edge to move across the frame, and it takes a CMOS sensor that same amount of time (if they had made it that fast) the results should be relatively similar in terms of the image.

We have no way at the moment to know any of the claims by both Panasonic OR Red. I'm a 'see it and believe' person as well. But I have the same skepticism with announcements from Panasonic as well. Especially the 'ND filter to remove Moire' bullet point I saw in their presentations. Which to me is clearly a disconnect between marketing and engineering.

For me, I can read '5ms read/reset' and compare that with the video and statistics of the RED ONE. This means for me it is significant. How will it compare to film? Who knows what other issues there are. But when *I* look at the math, it makes sense to me, and from the specifications and physics of how a film camera works, it tells me that we are getting closer. Testing will determine how close.
 
Barry_Green said:
Has anybody ever taken a Red with its 5ms sensor, and a film camera, side by side and tried to get jello out of them both? Or put long lenses on them and done a skew test?
Click to expand...

Blasphemy!

You should know better than that Barry. RED is god and nobody dares test it against any inferior (non RED) camera.

I dunno what the sensor read time is on the GH1 but it looks bloody good compared to the Canons. And now that it's mud free you can whip pan to your hearts content. That and the ability to use some cool exotic glass has got me thinking about adding it to the toy collection.
 
squig said:
Blasphemy!

You should know better than that Barry. RED is god and nobody dares test it against any inferior (non RED) camera.

I dunno what the sensor read time is on the GH1 but it looks bloody good compared to the Canons. And now that it's mud free you can whip pan to your hearts content. That and the ability to use some cool exotic glass has got me thinking about adding it to the toy collection.
Click to expand...

I actually contacted someone about using a 35mm and a Red One today. Basically, there is another project going on and I offered to add a little cash if they'd let me do some pan/tilt tests with their Red One and their 35mm cameras they are using. Who knows, I explained what I was after as I think A LOT of people would like to see this comparison. I just don't want to pay for a whole roll of film if I don't have to.

I have to say, I am pretty impressed with what is coming out of the GH1 with the hacks.

But, it still doesn't give the freedom to 'whip around' without horrible skew, case in point:

http://vimeo.com/6859995

I took stills from the video and it does look like they have very similar skew. I use a fixed length line in photoshop and rotated it till it matched in both. Was pretty much the same.
 
Hehe not flipping, still keeping the 5D. I can't really afford another body, too many mouthes to feed and with all the lens options the kids would starve. But I really like that 70Mbps+ mjpeg. However the 5D has been producing great stuff since the 24p update.
 
Well, see, that's what I'm talking about. I've never noticed film skewing or going rubbery, ever.

But, for that matter, most of us never noticed strobing in 24p until we started shooting our own... so yeah, I'm curious. If I knew someone with a Red now I'd just set it up, but how hard can it be for our L.A. folks to get that goin'?
 
Isaac_Brody said:
Saw this rolling shutter example for Red on vimeo, this is with the current firmware build. Shot by Ryan E. Walters.

http://vimeo.com/11095706

Film doesn't exhibit that kind of skew.
Click to expand...

But that is still 8.5ms read/reset. I want to see what 5ms read/reset looks like and I'd also like to see what film skew looks like as well. There is a great big giant lack of any examples on the web.
 
Barry_Green said:
Is that a serious question? Shao, where ya been? :D
Click to expand...

Haha, I wasn't sure how to take that question either at first! :) Do some searching on here for the firmware hacks for the GH1. They are really getting some interesting results with all the different things they are trying.
 
Swamped with life + new job.
Click to expand...
Really? So you haven't heard? Okay, check the date because it's *not* April 1st.

4:2:2. 80mbps MJPG. 1920x1080 MJPG. Mud-free 50mbps+ AVCHD. NTSC<->PAL switchability. 3rd-party batteries enabled. And lots, lots more. The GH1 is now being called the GH13, in honor of "tester13", the userID of the programmer who's been modifying the GH1's firmware. It's a whole different ballgame now.

And no, we're not lying. This is the real deal. The GH1 forum has gone *crazy* in the last week. The GH1 firmware tester's thread is now the #1 most-viewed thread in all the history of DVXUser.
 
Barry_Green said:
Really? So you haven't heard? Okay, check the date because it's *not* April 1st.

4:2:2. 80mbps MJPG. 1920x1080 MJPG. Mud-free 50mbps+ AVCHD. NTSC<->PAL switchability. 3rd-party batteries enabled. And lots, lots more. The GH1 is now being called the GH13, in honor of "tester13", the userID of the programmer who's been modifying the GH1's firmware. It's a whole different ballgame now.

And no, we're not lying. This is the real deal. The GH1 forum has gone *crazy* in the last week. The GH1 firmware tester's thread is now the #1 most-viewed thread in all the history of DVXUser.
Click to expand...

Sweet, that sounds awesome. I'll have to check it out.
 
Status
Not open for further replies.
Back
Top