AF100 / AF101 - initial report

[Barry_Green]

Rolf asks a hard question, and Stephen answers it in terms that I can largely agree with, but not entirely...

Here's the thing. It's hard to judge DVCPRO-HD against AVCHD when using the HPX170 as a source, because the chips in the HPX170 don't deliver all the sharpness that would truly tax either codec. It would be a more-even codec test to compare on something like an HPX3700, where the chips are delivering every bit of data that the codecs could stand, so hooking an HMR10 AVCHD recorder up to the output of an HPX3700, and recording native DVCPRO-HD on the HPX3700, would be the way to tell just what the differences are, in terms of what the raw codec can handle. Or, alternately, plugging an HPG20 (in DVCPRO-HD mode) into the SDI output of an AF100, would let you know how the chips and codecs act on that specific combination of hardware.

So -- is AVCHD a "step down" from DVCPRO-HD? In some ways, yes, in some ways no. It's 4:2:0 instead of 4:2:2. And it's long-GoP instead of intraframe. However, it's also 2001-era technology, vs. 2010-era technology. So would it be fair to say that AVCHD can deliver comparable images as compared to DVCPRO-HD? Possibly. An intriguing question.

DVCPRO-HD is 4:2:2, but it also prefilters the recording resolution down from 1920x1080 to 1280x1080. AVCHD is 4:2:0, but it retains the full 1920x1080. So DVCPRO-HD will have twice the color resolution but only 2/3 the horizontal resolved detail.*

And DVCPRO-HD is bulletproof as far as motion artifacting, whereas AVCHD is a long-GoP codec. However, as a long-GoP codec, it's really good; I compared it against XDCAM-EX and found AVCHD to be much more robust regarding motion artifacts from cases where the codec gets overloaded. So while it isn't a true intraframe codec, it's much less susceptible to the issues that caused long-GoP codecs to get the reputation they got in the early days of HDV.

The ultimate is AVC-Intra. Intraframe, high bitrates, 4:2:2, and 10-bit color. That would be the ideal. But that requires P2 and 10-bit infrastructure throughout the camera, and that's not the AF100. I'd love to have that camera (an AF150? AF200?) but it would cost substantially more (Panasonic has said between $2,000 and $4,000 more) and just how many of us are there who would really buy it? 5%? 10%? I bet 90% of the folks out there, when faced with the prospect of a $4699 AF100 or a $8500 AF200, would choose the AF100. And that's what Panasonic has to look at; building a unit that will satisfy the market.

So - am I concerned about it using AVCCAM instead of DVCPRO-HD? Not at all. Like Stephen, I have found AVCCAM to be a perfectly satisfactory alternative to DVCPRO-HD. I would still choose DVCPRO-HD for chroma key work, but for general work, AVCCAM holds up very very well. There's no doubt that the images from an HMC40 or HMC80 (in good light) are substantially more detailed than those from an HVX200/HPX170, and the AVCCAM codec is robust with little if any artifacting. I've been using the HMC40 and been very happy with the recorded images it produces. If I was making a $100k feature film, I might invest $2k into a Ki Pro Mini; if I was making a $500k feature film I might invest $5k to get an HPG20, to get the absolute maximum best-quality codec I could get, within the budget. But for a built-in codec I think AVCCAM is a totally reasonable choice. I think those who are complaining about AVCHD, are in for a big surprise when they start using it. Panasonic avoided the whole HDV thing, because they thought that HDV/MPEG-2 just wasn't adequate for professional use. They caught a lot of flack for that, but they stuck to their guns, and now they're betting heavily on h.264/AVCCAM, because they think that finally there's an affordable low-bandwidth codec that delivers professional-quality results. Give it a try. You might be very pleased with what you see.

Discovery HD approved DVCPRO-HD for unlimited acquisition, editing, and mastering. I don't know if they've given any rating for AVCCAM AVCHD yet.

*edit: when I say "twice the color resolution", that's in terms of the ratio of color to luma samples. But if you want to be strictly mathematical about it, the advantage isn't quite as great as twice, because you have to take the prefilter into account. So truly, DVCPRO-HD's recorded color sampling is 640x1080, as compared to AVCHD's recorded 960x540 color sampling. That works out to a 1.333:1 advantage in terms of total color samples. But even then it's not really quite fair, because you're talking about truly twice as many on the vertical... and then, furthermore, it gets more complex in PAL where DVCPRO-HD is 1440x1080 instead of 1280x1080 so the luminance resolution advantage of AVCCAM is not as great, whereas the color sampling advantage for DVCPRO-HD increases from 1.333:1 to 1.5:1. so... intriguing question.

 

[Rolf Schwarz]

thanx stephen and barry for the reply,feel much better now because
if the picture quality isn't an issue,i'm really looking forward to
get this cam,which is really compact compared to what i have now,puh...
one thing is missing,can i use the lcd monitor and the viewfinder to control
the sharpness for rack focus or do i have to use an external monitor?

 

[Barry_Green]

Well, whether the finder/LCD are adequate for rack focus is a matter of your own level of acceptance. On the prototype I used, I didn't know about/couldn't get the colored peaking feature to work, so I don't know for certain. If the colored peaking works the way that I discussed with the engineers, then I would say that it will be adequate for getting razor-crisp focus without needing an external monitor. However, without the colored peaking, I did end up frustrated on getting perfect focus on deep-DOF shots. It's easy to get perfect focus on shallow-DOF shots with the existing LCD, even without colored peaking, but deep DOF shots were a lot more of a challenge. The colored peaking should remove that last obstacle.

 

[Noel Evans]

If I was making a $100k feature film, I might invest $2k into a Ki Pro Mini[/SIZE]

I agree with that 100%, but I have some other reasons why the ki pro mini floats my boat.

Ive tested premiere pro with CS5 and its true AVCHD works in multi stream, then you go to AE. Reality bites. And I still prefer to cut in FCP. For me its not even so much about 10 bit. Whilst that would be great of course, Im more for the I frame 4:2:2. And the one simple reason is workflow. Im happy to invest $2k extra in the ki pro mini as Ill likely have had that returned in a short time - ready to edit clips dropped onto my system to edit drive and backup simultaneously.

For example, with the 5D now - import / log "good" clips / transcode / backup / import / edit OR deliver to client (depending on the job), and Id use much the same workflow with AVCHD.

Now I frame 4:2:2. Drop clips to edit drive - backup drive simultaneously / EDIT or deliver to client. PLUS I can also - drop clips to local and external HDD in the field and handover to client immediately. I can do that with AVCHD of course, but do both and then ask the client a week later which they preferred. SO for me its about time saving and service.

Of course, if you have plenty of time on your hands then you might as well save your money.

 

[Doctor Wu]

Did any of these prototypes / pre-production models have any kind of back focus adjustment?

It's going to get hard to get razor sharp focus pulling without that. Especially on zooms.

I know that on traditional 2/3" Panny cameras the back focus adjustment is on the lens, but this is not your traditional 2/3" camera. So Im wondering if "professional" calibration feature will ever make it to a working af100.

 

[Steve Kahn]

So -- is AVCHD a "step down" from DVCPRO-HD? In some ways, yes, in some ways no. It's 4:2:0 instead of 4:2:2. And it's long-GoP instead of intraframe. However, it's also 2001-era technology, vs. 2010-era technology. So would it be fair to say that AVCHD can deliver comparable images as compared to DVCPRO-HD? Possibly. An intriguing question.

And DVCPRO-HD is bulletproof as far as motion artifacting, whereas AVCHD is a long-GoP codec. However, as a long-GoP codec, it's really good; I compared it against XDCAM-EX and found AVCHD to be much more robust regarding motion artifacts from cases where the codec gets overloaded. So while it isn't a true intraframe codec, it's much less susceptible to the issues that caused long-GoP codecs to get the reputation they got in the early days of HDV.

The ultimate is AVC-Intra. Intraframe, high bitrates, 4:2:2, and 10-bit color. That would be the ideal.

I would love to see a comparison between AVC-intra or any intra codec and ACVHD like the one you did in the great XDCAM-EX-vs.-AVCCAM test. One could use the exact methodology with an AF100 (or any AVCCAM) and a ki-pro mini or nanoflash.

Now that would be interesting.

 

[Barry_Green]

Did any of these prototypes / pre-production models have any kind of back focus adjustment?

No, for many reasons, not the least of which is that cameras pretty much never have any form of backfocus adjustment. That's done on the lens, or through shims.

It's going to get hard to get razor sharp focus pulling without that. Especially on zooms.

Well, no, not really. As long as the sensor is the exact distance from the lens mount that it's supposed to be, that's all the camera has to do. Any further adjustment is done by the lens.

Besides, are you sure you're looking at this right? I mean, there are no ENG lenses that are going to be worth putting on the camera, because ENG lenses are designed for little sensors, so the zoom range on an ENG lens would cover way too short a focal range in order to be able to actually cover the sensor. So forgetting ENG lenses, then you'd be talking about SLR zooms, and SLR zooms are horrible -- they're not designed to hold focus at all. Even if they do hold focus (as in a parfocal design), their focus mechanisms are not in any way reliably calibrated. Simply put, they were never designed to execute a proper zoom. They're for stills, after all, and it's impossible to zoom during a still shot.

So the only lenses where this might be an issue, would be in actual cinema zoom lenses. And in cinema zoom lenses, you'll be stuck with the same thing that we face on cinema cameras - shimming the mount for proper collimation.

Besides, backfocus has nothing to do with pulling focus. You can set the backfocus on a Red One to a horribly misadjusted level, and still execute a perfect focus pull; you just wouldn't do so by measuring from a tape, you'd have to manually set the marks.

The only thing backfocus does is adjust the sensor-to-lens distance, and you do that on an ENG lens in order to HOLD focus throughout a zoom. It lets you zoom-in, focus, and zoom-back-out to shoot. Which is a nice thing, of course, but SLR zooms aren't typically designed to do that task at all in the first place. SLR zooms are frequently varifocal designs, which cannot (by design) hold focus throughout the zoom range.

I know that on traditional 2/3" Panny cameras the back focus adjustment is on the lens, but this is not your traditional 2/3" camera. So Im wondering if "professional" calibration feature will ever make it to a working af100.

Well, it has the same "professional" calibration feature that pretty much all cinema cameras have ever offered -- which is, none. The cinema camera's job is to provide an exact distance from the sensor to the lens mount. Any adjustment that needs to be done beyond that, is done on the lens.

Think about it, there's no backfocus adjustment on any SLR ever made, and I've owned dozens of movie cameras and not one has had any manner of backfocus adjustment. The only camera body I've ever owned that had a way to adjust backfocus on the camera itself, was the original Red One.

Adjusting the backfocus on the camera body would be a very perilous thing to do. If for some reason one of your lenses was "off", and you adjusted the mount's backfocus to accomodate that one lens... then you just made >every other lens you ever mount on the camera< focus wrong.

The mount should be exactly, precisely, perfectly a certain distance from the sensor, period. Adjusting that distance is a really, really bad idea, as a general practice. If you have some unusual and specific circumstance for adjusting it, well, maybe... but in general, no. Not without knowing exactly what you're doing, and having access to the proper equipment to ensure that you get it exactly right.

Backfocus adjustment ideally belongs on the lens, not on the camera body.

 

[Jaime Valles]

I would love to see a comparison between AVC-intra or any intra codec and ACVHD like the one you did in the great XDCAM-EX-vs.-AVCCAM test. One could use the exact methodology with an AF100 (or any AVCCAM) and a ki-pro mini or nanoflash.

Now that would be interesting.

Agreed. I'd love to see a test with the AF100 recording to AVCCAM PH vs Ki Pro Mini (and Atomos Ninja) to see how much better the image can be in 4:2:2. I'm sure Barry's going to be all over that ASAP.

 

[Barry_Green]

Agreed. I'd love to see a test with the AF100 recording to AVCCAM PH vs Ki Pro Mini (and Atomos Ninja) to see how much better the image can be in 4:2:2. I'm sure Barry's going to be all over that ASAP.

I'll definitely be conducting such tests. I want to know if there's a big improvement possible, or if it's even worth the effort, etc.

 

[Doctor Wu]

There's a huge difference between a regular FILM cinema camera and the way it maintains its flange focus distance and a CMOS sensor or any chip camera:

Chips are far more sensitive to heat variations which throw your back focus off.

That's why it's de rigeur for a camera assistant working with an electronic camera to set back focus every morning after the camera's been cold during the night and again around lunchtime. That's why Digi Primes come with a back focus collar where Ultra Primes do not.

My point with the 2/3" lenses wasn't that I'm trying to use 2/3" ENG lenses on an AF100.
It's that any professional CAMERA SYSTEM has a way to deal with back focus issues. Whether its on the lens, or on the camera (as in the RED).

Trying to use a Canon or Nikon or really any zoom lens on the AF100, without any back focus adjustment, will mean you'll never be working off the numbers. You'll always just be marking your disk. And you're WIDES may never match your TELE end with your back focus whacked like it is. I've been there many a time.

Shimming as a way of setting back focus on an electronic camera is a way to drive yourself crazy.

While that's perfectly OK, considering it's only a $5000 camera, it's not a professional way that camera crews and focus pullers work. I'll probably buy an af100 camera myself, but it's not without the reservation that you'll always be marking a disk, and probably always be dealing with strange back focus issues with zooms.

If there was a way for the AF-100 to set back focus automatically through some electronic means - that would be ideal, but I haven't heard of that feature from Panasonic or you yet.

But thank you for your coverage.

No, for many reasons, not the least of which is that cameras pretty much never have any form of backfocus adjustment. That's done on the lens, or through shims.


Well, no, not really. As long as the sensor is the exact distance from the lens mount that it's supposed to be, that's all the camera has to do. Any further adjustment is done by the lens.

So the only lenses where this might be an issue, would be in actual cinema zoom lenses. And in cinema zoom lenses, you'll be stuck with the same thing that we face on cinema cameras - shimming the mount for proper collimation.

Besides, backfocus has nothing to do with pulling focus. You can set the backfocus on a Red One to a horribly misadjusted level, and still execute a perfect focus pull; you just wouldn't do so by measuring from a tape, you'd have to manually set the marks.

The only thing backfocus does is adjust the sensor-to-lens distance, and you do that on an ENG lens in order to HOLD focus throughout a zoom. It lets you zoom-in, focus, and zoom-back-out to shoot. Which is a nice thing, of course, but SLR zooms aren't typically designed to do that task at all in the first place. SLR zooms are frequently varifocal designs, which cannot (by design) hold focus throughout the zoom range.


Well, it has the same "professional" calibration feature that pretty much all cinema cameras have ever offered -- which is, none. The cinema camera's job is to provide an exact distance from the sensor to the lens mount. Any adjustment that needs to be done beyond that, is done on the lens.

Think about it, there's no backfocus adjustment on any SLR ever made, and I've owned dozens of movie cameras and not one has had any manner of backfocus adjustment. The only camera body I've ever owned that had a way to adjust backfocus on the camera itself, was the original Red One.

Adjusting the backfocus on the camera body would be a very perilous thing to do. If for some reason one of your lenses was "off", and you adjusted the mount's backfocus to accomodate that one lens... then you just made >every other lens you ever mount on the camera< focus wrong.

The mount should be exactly, precisely, perfectly a certain distance from the sensor, period. Adjusting that distance is a really, really bad idea, as a general practice. If you have some unusual and specific circumstance for adjusting it, well, maybe... but in general, no. Not without knowing exactly what you're doing, and having access to the proper equipment to ensure that you get it exactly right.

Backfocus adjustment ideally belongs on the lens, not on the camera body.

 

[mainstreetprod]

Ok, somebody 'splain this: I've been involved in video since there were single TUBE cameras (used 'em in college with a U Matic recorder slung over my back). Eventually pro gear transitioned to three tubes, then three chips, because that was the only way full depth of color could be achieved. It stayed that way right up until the DSLR craze (I've never shot anything but a still with those so wouldn't know..) So for the last 30 years it has taken 3 sensors for great color, with consumer cameras having one chip and flat color. Now suddenly one large sensor achieves equal color rendition? How?

 

[Jaime Valles]

I'll definitely be conducting such tests. I want to know if there's a big improvement possible, or if it's even worth the effort, etc.

Excellent. I'd be interested in stuff like color bleed on bright red subject matter, how quickly the image falls apart during color grading, resolution difference (if any), and performance in front of green-screens. In theory the Ki Pro Mini 4:2:2 ProRes footage should be considerably better than the 4:2:0 AVCCAM, but not necessarily $2000 better (not including cost of CF cards!)

Definitely looking forward to your test results!

 

[Matt Gottshalk]

Ok, somebody 'splain this: I've been involved in video since there were single TUBE cameras (used 'em in college with a U Matic recorder slung over my back). Eventually pro gear transitioned to three tubes, then three chips, because that was the only way full depth of color could be achieved. It stayed that way right up until the DSLR craze (I've never shot anything but a still with those so wouldn't know..) So for the last 30 years it has taken 3 sensors for great color, with consumer cameras having one chip and flat color. Now suddenly one large sensor achieves equal color rendition? How?

Because that one sensor has all the color sensors, RGB, within it.

 

[BobDiaz]

Ok, somebody 'splain this: I've been involved in video since there were single TUBE cameras (used 'em in college with a U Matic recorder slung over my back). Eventually pro gear transitioned to three tubes, then three chips, because that was the only way full depth of color could be achieved. It stayed that way right up until the DSLR craze (I've never shot anything but a still with those so wouldn't know..) So for the last 30 years it has taken 3 sensors for great color, with consumer cameras having one chip and flat color. Now suddenly one large sensor achieves equal color rendition? How?

I'd hate to think what a beam splitter for a 4/3" image sensor would cost. However, with the 4/3" image sensor, the chip is higher than 1920x1080 pixels. There's no official statement from Panasonic as to the native resolution for the area covered by the video, but assume that's it's 3,840h x 2160v. (It's higher than that, but this makes my example simple.)

Each of the 1920x1080 pixels has 4 separate pixels on the image sensor to cover its area: 1 red, 1 blue, and 2 green. The AF-100 has an optical filter that lowers the image sharpness to match the 1920x1080 resolution, so every point on the 1920x1080 image has a minimum of one of each color. Thus, no need for the expensive 4/3" beam splitter and 3 separate image sensor chips.


Bob Diaz

 

[Rick Burnett]

Well we don't even know for sure if they are using a Bayer pattern, since they are not using the full resolution of the sensor, who knows that they are doing with the filter layer. It's possible they could use some variant with however they are doing pixel binning. For instance, using say 2 green pixels under the same lens can allow you to reject noise more since you know they should have the same value.

Of course, the really interesting stuff is about our eyes sensitivity to green, and how we sample twice as much green as blue and red. There is a lot to read online about it that is fascinating.

I will DEFINITELY be at least renting a KI Pro mini for some VFX testing. We'll be doing a lot of blog posts as we compare the AF100 to the 7D as we've shot so much with the 7D, we know EXACTLY what we don't like

After Barry's initial report with the prototype, I've got everyone excited. Moire and Rolling shutter being the 2 huge things.

 

[Michael Olsen]

Ok, somebody 'splain this: I've been involved in video since there were single TUBE cameras (used 'em in college with a U Matic recorder slung over my back). Eventually pro gear transitioned to three tubes, then three chips, because that was the only way full depth of color could be achieved. It stayed that way right up until the DSLR craze (I've never shot anything but a still with those so wouldn't know..) So for the last 30 years it has taken 3 sensors for great color, with consumer cameras having one chip and flat color. Now suddenly one large sensor achieves equal color rendition? How?

Well, it's not quite apples to apples.

A 3 chip camera recording from a beam splitter could theoretically record 100% of the color information in any given scene.

A single chip bayer pattern (most likely what is in the AF-100 along with the DSLRs, RED, and Alexa, etc) contains photosites which are filtered to record light in a R G B G R G B G pattern. As such, it can theoretically record 50% of the Green, and 25% of the Red and 25% of the Blue in a scene. This can then be processed and interpolated to make up for the data which was unable to be actually recorded. In other words, the camera (or the debayering software) is making educated guesses and filling in some blanks. To make up for this lack of color resolution, Bayer sensors are generally downscaled. Arri's D21, for instance, is a 2880 x 1620 pixel full sensor, but is down scaled by 1.5 for 1920 x 1080 final product. Similarly, RED Epics shoot on a 5K sensor, but output a 4K file.

And, for what it's worth, the DSLR for video craze wasn't really the groundbreaking use of bayer CMOS sensors...that was DSLR for stills!

 

[Barry_Green]

There's a huge difference between a regular FILM cinema camera and the way it maintains its flange focus distance and a CMOS sensor or any chip camera:

Chips are far more sensitive to heat variations which throw your back focus off.

Understood and agreed. But it still doesn't change that it is nearly unheard of for a camera itself to have sensor-to-mount adjustment. That's always been done on the lens.

 

[Barry_Green]

Ok, somebody 'splain this: I've been involved in video since there were single TUBE cameras (used 'em in college with a U Matic recorder slung over my back). Eventually pro gear transitioned to three tubes, then three chips, because that was the only way full depth of color could be achieved. It stayed that way right up until the DSLR craze (I've never shot anything but a still with those so wouldn't know..) So for the last 30 years it has taken 3 sensors for great color, with consumer cameras having one chip and flat color. Now suddenly one large sensor achieves equal color rendition? How?

Nothing has changed. Three chips still deliver superior color to a single-chip system. The reason everyone's going with a single chip is NOT increased color fidelity; a Bayer chip is inferior to three chips for color fidelity, and that's why every broadcast camera on the market has three chips and a prism. But there are some things that just can't be done with three chips -- primarily, emulating the way a film camera works and the ability to use lenses that have been developed for film cameras. A single sensor directly emulates the way a film sensor works, and you don't need to make accomodations in the lens design for a beam-splitting prism.

If you want to make a film-emulating camera with interchangeable lenses and the ability to take film lenses, you basically are locked into using a single sensor (whether Bayer, Foveon, or RGB Stripe).

 

[Jim Brennan]

I'm not Barry, but…

.

If I had a nickel for every time I've said that....:grin:

 

[ng smith]

Understood and agreed. But it still doesn't change that it is nearly unheard of for a camera itself to have sensor-to-mount adjustment. That's always been done on the lens.

Maybe for chip cameras, but PL mount film cameras are adjusted with shims on the back of the mount to camera and I would expect the Arri D21 and Alexa to be adjusted the same. Red camera adjusts on the mount. Maybe a way to adjust PL mount lenses on the AF100 is to incorporate use of shims or a screw cam in the adaptor? If that would be possible.