Barry,

You can't request Intraframe, and bemone the fact that DSLRs don't use B-frames.

The thing about high-end h.264 implementations is that they can have very high compression ratios, but they are computation intensive. DSLRs need to run for hours on a small battery. Using a simple implementation and high bitrate is the most effective solution.

Long GOP is fine for acquisition. But it stinks for editing. So we transcode. This has little to do with quality. It's a workflow thing.

The real limitations are 4:2:0 and 8-bits. Nail those exposures, people!

"Mediocre quality from the h.264 format" suggests that the footage shows codec-related issues so severe they make the image quality "mediocre".
Please show me a piece of original, ungraded 5D/7D footage (the GH1 is off-topic - see thread title) that shows these codec-related issues.

That doesn't make ANY sense. With a high enough bitrate, those b-frames are pointless. The entire reason to use b-frames is to get the bitrate low enough and for it to still look good. That's why AVC-Intra is only keyframes. It doesn't make sense to praise AVC-Intra as the best codec in existence (It's just the I-frames) and then say that the coding mechanism in the 7D is bad because it doesn't have b-frames.

The less motion estimation there is, the better the image. That's why they don't use b-frames when the bitrate is high enough. That's why AVC-Intra is only keyframes.

Download the raw video midway down the page:
Canon EOS 7D Hands-on Preview: 14. Samples: Digital Photography Review

Every frame of 5D/7D footage is full of codec related issues.
Whether people have the eye to see it is a different question.

It probably depends on the quality people are used to looking at. Any colorist would call the footage complete garbage from a technical point of view.

It's better than an HV30, but not as good as a prosumer camera like the HVX. But the HVX wont do shallow DOF without an adapter, to there's a trade-off.

I agree that the particular clip you linked to is not the greatest piece of cinema, but can you say exactly what the "codec" issues are that you see in every frame of 5D/7D footage?

Are you saying these issues are things that prosumer cameras do not suffer from?

Am I also detecting a Panasonic vs the rest of the world subtext to this thread? I don't know if we're descending dangerously close to an unquantified "favourite brand" mud-slinging match?

Such as? I don't see blocking or codec-induced "crawlies" in any of the motion or anywhere else. And it's not the codec's fault that the highlights are clipping.

I think you're going to have to point out at least one codec-related issue on a screengrab because I don't see any. Does anyone?

P.S.: Several frames blur out significantly because the camera is shaking. Nothing to do with the codec there either.

This discussion is "deja vu all over again." When professional HDV cameras came out, people trashed the codec, yet it's used by thousands quite successfully and even on the big screen. It is true that you can't push it as much as some other codecs--which means you have to be professional in your exposure and white balancing; you can't be sloppy as you can with different codecs. Presumably H.264 will be similar. In the hands of people who know what they're doing, it will be fine. You under or over expose something, you can't do as much correction as you can with something else.

I don't really think you can compare these things on the web. You can get all sorts of artefacts from the distribution network that have nothing to do with the original material.

My experience is that fine moving objects like leaves and grass tend to throw things with the codecs.

Regarding the 5DII on the big screen, the comments I've read on the BSC evaluation varied between "interesting" and "awful", it was the EX3 that impressed. Perhaps that was early days and there has been the objection was that the camera was used straight out of the box, so possibly the jury is still out in that regard. Although, the resolution figures referred to in another thread do tend to suggest that big screen detail won't be its forte.

The link that was provided appears to be an original .MOV file straight out of the 7D, so I don't object to using this to judge the quality of the codec.

Do you have a link? While aliasing is my big (and only major) caveat with the 5D, I was very surprised to find that properly focused clips contain a ton of detail.
What I did was extract screengrabs from various videos I shot. Then I scaled these down to DVD resolution and scaled them back up. There was a massive difference, no huge surprise there.
But when I scaled them down to 720p and back up to 1080p, I still noticed a difference in detail. So I concluded that it must resolve more than 720p. Probably not a true 1080p but if it's somewhere between 720p and 1080p then that's a pretty good result I'd say.

If you consider aliasing to be detail, then you're right, the 5D2 clips do contain a ton of "detail".

But if you don't consider aliasing to be a valid part of "detail", then the story comes out much differently. For example, in a normal 1080p camera, you can measure the resolution in the form of line pairs per picture height (lp/PH), from 1 lp/PH on up. At some point, you will start to see aliasing. In my ideal 1080p camera, it will go all the way to 360 lp/PH with no aliasing at all. On other cameras, aliasing starts at around 300 lp/PH. On the 5D2, the aliasing starts at a *very* low resolution. I haven't measured it but I'd guess under 200 lp/PH.

Almost any camera blows the 5D2 out of the water when it comes to alias-free resolution.

Resolution test...


About 800v x 700h ish (is what another poster on another board said)? But I couldn't read what this chart means for the life of me. And no aliasing on the diagonals!? Can see moire patterning, though.

I prefer some real information like what Alex posted. His Vimeo link says:
"Resolution measured @ approximately 1200x900 pixels"

This seems reasonable, and it's "a tiny bit" more than 200 lp/PH.

I'm sorry that you were not able to understand the information I posted. I'll try to put it more simply.

• There are many ways to measure resolution. For one, you could completely ignore aliasing and just measure the maximum number of lines possible. With such a ridiculous methodology, you could, for example, measure 1500 l/PH in the 7D test linked above.
  http://thebrownings.name/images/2009...mpets-crop.png
• Second, you could measure resolution at some arbitrary and subjective level, where there is still a strong aliasing, but it is amount that you, personally, find acceptable.
• Third, you could measure the resolution at which there is no aliasing.

The point is that the first and second are very far from the third in DSLR video. In high quality video cameras they are very close together. It's so bad in the 7D that the aliasing actually went off the chart. It only went down to 500 l/PH, but the aliasing was still very strong. You need a chart with far lower resolution in order to measure just how bad the 7D is.

So again, if you consider "aliasing" to be "detail", then the 7D can hit 1500 l/PH. But if you measure the amount of alias-free resolution, it's somewhere well under 500 l/PH (166 lp/PH if you use a Nyquist of 3 lp/PH).

I understood you fine. According to your definition of "detail" (a resolution with no aliasing), these cameras are about SD quality. Not a lot of people will agree with this.

Yes.

You're right. I'm part of a small minority of people who dislike aliasing very much. It so happens that most video camera manufacturers are also in that minority, but perhaps video DSLR will change that.

I'm also in a minority about many other problems that commonly affect any video camera, but "not a lot of people" consider them problems: oversharpening, posterization, poor color accuracy, interlace twitter, oversaturation, compression artifacts, missed focus, clipped highlights, crushed shadows, chromatic aberration, harsh bokeh, distortion, camera shake, and many more. Fortunately, my 5D2 does far better than most cameras in most of these areas (with some notable exceptions).

I don't know what percentage of people would agree with me about those things either. Probably very few. But to me, aliasing is just another one of those problem areas, although I dislike it more than all the others.

I dislike aliasing as much as other artifacts, but it just doesn't occur all that often in real-world applications.

To me, the 5D/7D have too many advantages for this to be a killer argument. And if you think about it, how many artifacts do we put up with when we watch anamorphic feature films? Blue streaks on light sources, ugly oval shaped Bokeh, heavy barrel distortion.

I watched Lymelife recently, an independent movie with Alec Baldwin shot on anamorphic 35mm. There's one scene where two women talk in a supermarket. One is on the left of the frame, the other is on the right of the frame. While the center of the frame was sharp, both women were quite blurred and showed clearly visible chromatic abberation. Poor lens performance towards the sides. How many people will even notice let alone be bothered by it?

I've shot a lot of 5d material and there has been NO shots that have been ruined by aliasing. And the resolution is absolutely wonderful. Way better than any 35mm adapters with regular videocameras like the sony ex-1 could ever hope to achieve.

Spec-O-mania!
 

Reality check:

How many people in this thread have honestly EVER had their material shown theatrically? We should do an anonymous poll.

It always cracks me up how almost everyone on all of the sites I frequent go on and on about how their camera must have XXXX and YYYYY and ZZZZ codec and features so that the image will look good on the "big screen", yet 99% of their work is only seen by an audience on the web or SD DVD. Almost any camera on the market today, through the use of good lighting, composition and movement, can be made to look amazingly good for the web or SD DVD.

Generally, if you have the money for a filmout, or an HDCAM or HDCAM SR master for 1080 projection digitally, you usually have the resources to shoot on a more sophisticated camera than a $1,700.00 DSLR anyway or at least you should.

720 vs. 1080
8bit vs. 10bit
4:2:0 vs. 4:2:2 vs. 4:4:4
Long GOP vs. intraframe
native resolution sensors vs. pixels shifted
RAW vs. processed images

None of this stuff should REALLY matters to 99% of the people who pontificate endlessly about it. There are a people that this stuff should matter to. But the vast majority of them are not shooting with DSLRs and prosumer cameras and are not on the web debating endlessly about it.

Just my .02.

Dan

Well, I have and lots of others have, not just at festivals but often for small theatrical runs, as well as on various TV networks in the U.S., Canada and Europe. Anybody who knows what he's doing can work within the limitations of most any professional camera and make the footage look good. Not as good as the high end gear under the same conditions, but good enough to work. Hell, look at those guys who did that disgusting Crank 2 movie shot mostly with little single chip AVCHD cameras. They got shots they couldn't have got any other way and they looked good. Unfortunately the film was putrid, but teenage boys like it.

Footage from the 5D is cropping up everywhere even on network shows, people are using it on Steadicams to intercut with 35mm footage. No, the 7D is not going to be a Red killer, and nobody who knows anything would suggest that, but it looks a hell of a lot better than many other cameras costing a lot more and in the hands of good cinematographers is going to be a valuable tool. I've even seen footage from that little Lumix GH-1 that looks better than HVX200 footage, even though it has a codec people call amateur. Who cares, if a talented person can make good movies with it. Nobody's making anybody go out and buy an HDSLR. People who have bigger chip "real" video cameras aren't going to dump them for a hybrid. These cameras are like any other cameras: another tool.

Something that seems to be forgotten is that not everyone wants shallow DoF. Can you imagine what it would be like trying to cover a sports event with long lenses and ultra shallow DoF. How about racing, where only the leader is in focus, assuming the camera operator is good enough to pull focus on moving object. The majority of television applications need deep depth of field. Large sensor, shallow DoF wont work on chat shows where you want to have all the guests in focus at once. any of the most breathtaking shots seen on TV rely on deep DoF, like the opening waterfall shots from the BBC's "Planet Earth" or the vast herds of cattle roaming the Savannah.

Sure there is a place for shallow DoF, but I think it is a fad that is currently being overdone. Next time you watch a theatrical release film look at how few supper shallow DoF shots there are. For these reasons I don't believe that large format sensor camera will ever replace 2/3" cameras for the majority of productions. Give me a well sorted 2/3" camera with proper optical low pass filtering tailored to the recording resolution. I don't want to go back to the days of making sure any on screen talent doesn't wear clothes with patterns or stripes for fear of moire and aliasing.

I'm sure we disagree about what constitutes aliasing. Most people think the only aliasing artifact is moiré, which I agree does not occur that often in real-world applications. But there are many other aliasing artifacts. I've been shooting my 5D2 for almost a year now, and with other 5D2 cameras in weddings and multi-cam shorts, and I've seen lots of footage posted online, but I have never yet seen one .MOV clip from a video DSLR that had both fine detail and no aliasing. Most of the time there is no fine detail because the DOF is too thin, focus was missed, subjection motion, soft lenses, diffraction, etc.

When I have stated that in the past, people are quick to defend the 5D2 and post many links to Vimeo that they say only has a "little" aliasing, here and there. In fact they are riddled with aliasing all over, but only have a little moiré, here and there. So it's obviously a difference in perception.

Here is one way to help illustrate the difference. The following image is riddled with aliasing artifacts, just like the 5D2:

http://thebrownings.name/photo/2009-...-400-point.png

And here is the same image, but with no aliasing artifacts:

http://thebrownings.name/photo/2009-...00-lanczos.png

Some people are not be able to see any difference between the two. Many would prefer the first image, describing it as sharp, crunchy, high microcontrast, with lots of fine detail, such as stubble. Others would see it as fake-looking, with harsh transitions, jagged edges, and lots of false detail, such as stubble that should be too small to see and jagged edges on the ear.

Some would prefer the second image, describing it as smooth, natural, with the appropriate amount of detail for its size. Others would see it as mushy, hazy, low contrast, and lacking in fine detail.

Which image do you prefer?

Can you see how that would result in a lot of miscommunication? When I say I dislike the first image very much because of aliasing, many people would respond by saying that they don't see any aliasing in it, and in fact they prefer the first image.

Others agree that there are indeed aliasing artifacts, but that you have to really pixel peep to see them, so the overall image is not affected. I highly disagree. To me, the aliasing causes the overall image to take on a very fake, computer-generated look. For me it's just as jarring as the difference between 24p and 60i: I notice the difference in the overall image, not just when pixel-peeping.

Agreed. Sometimes the low light and/or DOF advantages trump all else. Some scenes are so dark that my XH-A1 would not get a usable picture at all. When you compare an aliased image vs. no image at all, the aliased one wins every time. Furthermore, many audiences will prefer an aliased thin DOF image over a clean, high-res deep-DOF one. That's another good time for me to reach for my 5D2.

Good example. I'm sure many (most?) people actually prefer aliased images over anti-aliased, just as many people like the anamorphic artifacts. I like the oval bokeh and blue streaks.

It's not just specific little artifacts, it's the overall feel of the image. For example, say you ask someone to choose between two styles. The first is way oversharpened, oversaturated, over-contrasty (blown highlights, crushed shadows), heavy noise reduction, aliasing, and 60i. The other has low sharpening, neutral saturation, low contrast (no blown highlights or crushed blacks), no NR, anti-aliased, and 24p. Many people will highly prefer the first. That's why digicams are set out of the box to provide that kind of image: marketers know that it sells.

Your personal taste does not include a dislike of aliasing, and that's fine. So when aliasing occurs in your footage, it does not ruin it for you.

"Better" is subjective, even in the hands of skilled users. For some viewers, thin DOF and low light performance is the be-all end-all. For them it wouldn't matter if the 5D2 had banner ads embedded on the recorded video (somehow licensed so they could not be removed or hidden in post), they would still prefer it over cameras that had deep DOF, and would just assume that their audience will not mind the banner ads (perhaps correctly). For others, banner ads would be a deal-killer, no matter what the other benefits. To them, thin DOF is not worth the distracting banner ads.

In the same way, aliasing is distracting and jarring like banner ads. To them it is a much more difficult choice. One has thin DOF, low light, and terrible aliasing, whereas the other has deep DOF and no aliasing.

Agreed.

Agreed. Lots of top cinematographers, like David Mullen, prefer to shoot f/4 or f/5.6 on S35 whenever they can get a big enough lighting budget. That's equivalent to f/2 or f/2.8 on 2/3" and f/6.3 and f/9 on the 5D2.

I prefer the second image because it looks more natural BUT, and this is where our points of view differ, I don't consider the aliasing in the first image a show stopper at all. If you hadn't posted the second image, I'd find it difficult to even pick up on the artifacts unless I scrutinized it very closely.

Also, both images are 400x500 pixels in resolution (let's call this HD). If you now scaled the aliased image down to half that (let's call this SD) and back up again, it would end up looking a lot worse to me. Whereas you would probably find the aliased HD about as bad as the SD.

That's it in a nutshell. To some, the artifacts are no big deal at all, but to others, it's a very jarring difference, like 60i vs 24p.

I don't see that much aliasing with the 5DII. At least with sharpening turned down. The second image seems pretty soft for the 5DII.

Interesting. I think we have different perceptions of the same image. I'm curious what you think of the following comparison between the 5D2 and HV10:

http://www.dvinfo.net/forum/1132912-post31.html

The simple fact is that nobody is going to stop the video and look closely at a single frame when watching a movie - well, not 99% of the viewing public.

What matters most is how it looks when the footage is running, and when it is viewed at an average distance from the screen.

You have it backwards. If these were still images, aliasing would not be such a tremendous problem.

When you're looking at just one frame or a still image, the aliasing is actually not that bad. But when the footage is running, that's when aliasing goes nuclear!

Still frame:

• Aliased white specks in hair might be light glinting off hairspray.
• Jagged edges could be that way in real life.
• Stair-stepped lines could really have different widths.

Video in motion:

• White specks start dancing and popping, like fireflies in their hair.
• Jagged edges start flowing like waterfalls.
• Stair-stepped lines start changing width and popping in and out of visibility.

Aliasing in a still image is bad enough, but it is 100 times worse in motion.

100 times worse in motion? I had to laugh at that one :)

I'd also have to disagree with you on most points, Daniel, and more inclined to agree with Cris and Don.

Although tight registration is the issue you're concerned about, it's not those sorts of patterns you should be concerned about.

It's literally the threads of fabrics that are actually resolving on the camera's huge sensor. I know this because I see them in still shots where the focus is dead-on. But in video of the same exact lens/focal length/settings, the video moires.

The same goes for distant chain link fences, brick buildings or patterned roofs.

There are ways to hedge your bets, but they are all pretty experimental in nature at this point. I'm personally thinking about adding a softening filter for all shots. Still not sure exactly sure about which one....

While that's certainly true, the aliasing problems we're discussing are a little bit more obvious than that. I think any professional would find them objectionable.

I don't know that it's exactly 100 times worse, but Daniel is spot on here.

If you look at the first photo alone, you won't necessarily notice aliasing. Look at it side by side with the non-aliased photo, and you can find the aliasing, if you're looking for it. But this isn't reality. When somebody shows you a photo, they don't usually also show you the less crappy version to compare.

However, if the aliased image moves slowly, the over/under done highlights in the hair will start jumping around. They draw the eye - and you don't need a 2nd photo to find the problems.

So, yeah, aliasing is way worse with motion than stills.

That 5dII image looks pretty bad.

I do see a lot of aliasing in HD cable TV.

So how come we've seen all that beautiful footage from all those 5DMKII films if it's so bad?

There are plenty of shots that show zero aliasing. But when it sneaks in, you have to be ready to deal with it - often instantaneously. The quick fix is to defocus the shot slightly. This is a risky enough maneuver with the camera as the DOF can often be very shallow.

I find myself generally lighting and setting ISO to get the aperture somewhere around 5.6 in order to widen that focal plane a bit. That also will make more out of my zoom lenses - which are 2.8 L series glass. So, a nice side effect, that.

But it does make for harder-to-control aliasing. I'm still thinking a slight softening filter will help with various subjects. I need to get my hands on a few different densities and types to really test the theory, though... Anyone with a killer filter set want to report?

What type of optical low pass filter do these cameras have? There could be a mismatch between the stills requirement and the video.

First of all, the 5D2 has lots of good things going for it. Even if someone dislikes aliasing, those other things may be more important and/or make up for it. Second, there are lots of people who like aliasing, so for them it actually improves the image instead of making it worse.

Let me illustrate. The 5D2 allows control over the sharpening, contrast, and saturation. But imagine if it had no such control, but instead had all parameters cranked up to the top. For some people, that would be fine, they like the look of oversharpened, blown whites, crushed shadows, and hyper-saturated colors. Others would greatly dislike it.

I agree that there are plenty of shots that show zero moiré. And that there are plenty of shots that don't have enough fine detail to show aliasing. But I have seen very few shots that have both fine detail and zero aliasing. Probably part of the difference is that you and I may look at the same image and where I see aliasing, you might not see it. (Like the example image I posted above.)

Canon uses two layers of lab-grown, ground, and polished Lithium Niobate crystal, like so:

Sensor-Film: Low pass filter

They have been making the blur smaller, relative to pixel pitch, with newer cameras since the 30D. (That is, MTF curve of the OLPF has moved to the right, relative to Nyquist.) This has made aliasing in still images slightly worse.

Most certainly. The 30D OLPF was about right for my taste, but the newer cameras (50D) introduce a little too much aliasing even for stills, based on the results of Siemens Star tests. Of course, they are an order of magnitude too weak for row-skipping and binning 1080p video. Even for downsampled video they're too weak, but only by a few tens of percent in linear resolution.