Canon EOS 5D Mark II Review Discussion

[dougthebiker]

How does the 5DII perform for sports? Does anybody have any quantified data on the % of out-of-perfect-focus shots you get in a football or soccer game?


And how does that compare to the % OOPF of a 1DIII or a 50D?


Thanks for the help. I'm trying to decide what body to buy next. I can live with the relatively slower frame rate of the 5DII, but I'd want a "close to" 1DIII autofocus performance to make up for that. I'm not wild about the IQ of the 1DIII and I understand the OOPF of the 50D is pretty weak for sports.

[Mr Chad]

(Video on the 5D II)


I just wanted to add for those running Apples. I'm using a 15.4" MacBook Pro (2007) with a 2.2gHz Core Duo and only 2GB of ram. I'm able to view and run video from the 5D II without any issues (in Full HD settings on the camera). I have not yet however, attempted to edit those same files in any large way. But they do play just fine as .MOV files on a year old or so entry level MacBook Pro. I was really surprised I was able to do this.


I believe Canon says a 2.6gHz is recommended for Apples. Apple's computers tend to be fairly fixed in their configurations, so this may be of use to anyone running an Apple as a data point for running video. I thought this might be useful to some, since Brian mentions his experience with Dell laptops in the review.

[Oren]

As far as I know, you don't need a strong computer at all to play HD movies... not only that, but you can probably play them with a pretty old/weak computer.


I really don't know what Bryan was referring to, and his statement really got me confused - since as far as I know, pretty much any computer can play an HD movie - let alone a 2.5GHz Core 2 Duo with 4GB RAM.


It's not like a game where there is a sophisticated rendering work for the computer and graphic card or like an editing tool which makes intensive and complicated computations as far as I know.


Correct me if I'm wrong.

[peter buehner]

I don't own the 5D II but I use a 1D III and 40D regularly. I don't think the 5D is where you should look for sports. As mentioned in the review, the shutter lag, slower focus, limited AF points, and slower shutter speed, are all deal breakers.


The 40D is ok


The 1D mark III is simply fantastic. You get what you pay for.


If sports is only going to be a part-time thing and the rest portrait, studio, or landscape...well then I think the 5D II would vault to near the top of the pack.





good luck,


Peter

[tlinn]

Bryan,


Thanks for the great 5D Mark II review. You mentioned the Rode Stereo Video Mic and implied that it was relatively small. Would it be possible to post an image of the SVM on a 5D2? I know this mic is highly regarded but from the images I've seen -- all on camcorders -- it doesn't look that small. I'd love to get an idea of it's size relative to the 5D2 body.


Thanks.

[tlinn]

• Stop warping the histograms by 1-3 stops (i.e. use raw instead of jpeg).

I wonder if it would be more accurate to complain that the histogram doesn't reflect 14bit data when shooting raw? You can't really display the linear RAW data on a histogram and make sense of it because it has to be run through a tone curve. The inaccuracy of the histogram results from the additional headroom that results from 14bit data that is not accounted for in the supplied histogram, along with the fact that the tone curve you use to "develop" the RAW data will be different than the one the camera uses to create the histogram. This is why the histogram changes when you change picture styles, even though picture styles don't affect RAW data (i.e., the camera is using a different tone curve to "develop" the histogram from the RAW data depending on the selected picture style).

• Stop deleting up to 5 stops of highlights from the RAW data (i.e. allow ISO to be metadata).

Perhaps I'm just not getting what you are trying to say here. Canon is not throwing away 5 stops of highlights, although I can envision a scenerio where one could conclude this based on how one might imagine the ISO function of a DSLR works. A sensor has only one actual sensitivity. Put another way, eachpixel on a sensor captures so many electrons during a capture.For any given shutter speed and aperture there is no way to make it capture more or fewer of them.Any ISO above the base ISO is simplythe result of increasing the gain, amplifying the actual signal (and noise).


For the"L" and "H" settings,the increased sensitivity isn't even the result of an amplication. It is a basically a software manipulation. In the case of images shot in one of the H modes, they are basically ISO 6400 shots with a metadata notation (in the case of RAW files); likewise, L shots are ISO 100 shots with a metadata notation.


Of course, you may have a completely different rationale for making this claimto which I would simply suggest that no company is going to delete 5 stops of highlight data if thereis an easy way to avoid doing so. Dynamic range is becoming the new grail in the DSLR world and everyone (except Fuji) is stuck in roughly the same place. If Canon was too stupid to take advantage of a way to squeeze five additional stops (!) out of their sensors, someone else would.

• Stop clipping highlights in raw files taken at f/2.8 or less just to compensate for sensor AOR.

I'd be interested in a more detailed exposition of this point. I've never read this claim and I'm not sure what you mean by "sensor AOR".

• Stop ignoring masked-off pixels: use them for something, such as removing variable pattern noise.

Interesting idea.

• Stop crippling the Auto ISO for the very purpose that it is most useful: manual mode.

Amen. Amen. AMEN! WTF, Canon?

• Stop hiding important features such as the dead pixel remapper and shutter count.

Doesn't this seem obvious? Olympus was offering pixel remapping on their digicams already in '04.

• Stop wasting huge amounts of storage on ludicrous compression like sRAW.

I won't use this mode either but many folks will -- including alot of pro wedding photogs.

• Stop overriding the photographer's exposure settings just because it's in video.

What gets me is that, from what I can tell, the photog's settings are also overridden for stills taken during video capture. I'm not completely convinced that this is simply a case of Canon withholding features (like a real auto-ISO) to segment the market.

• Stop limiting video to 12 minutes; start a new file at 4 GB.
• Stop pumping the audio gain up and down constantly: let the operator set it.
• Stop forcing the use of the most incompatible framerate (30p) in the world.

This is a commonly-raised objection but it is misleading. For users in the United States, 30p is fully compatible with the ATSC high def standard and playable on all high def TV's and any computer monitor. It can also very easily be downsized to standard def NTSC 60i or converted to 24P HDTV.


The real issue is the cinematic look you get with 24p. This isn't a compatibility issue as much as an aesthetic one and for folks wishing to insert 5D2 footage into a film with other 24p footage it can be an issue insomuch as the look of the video will be different. On the other hand, 30p has 25% more frames than 24p which will help reduce strobing when folks pan while shooting video. I would be willing to wager that this is a bigger benefit for most 5D2 users than the cinematic look of 24p.


Most people will be VERY pleased with 30p. Of course, ideally you should be able to choose.

[ultima16888]

someone's angry. =P i was under the impression that sRaw is not a compression but simply a down sized raw file (less megapixel)

[Daniel Browning]

Thank you for the response. I enjoy the discussion and I am happy to go into more detail about the points I mentioned above.

You can't really display the linear RAW data on a histogram and make sense of it because it has to be run through a tone curve.

You're right about a linear histogram: it would be pretty useless. Maybe someone would appreciate it for a closer look at the exact distribution of the brightest 2-3 stops. I would not use a linear histogram. To simulate human vision, the histogram must have a gamma (or tone curve, as you said) of some sort.

I wonder if it would be more accurate to complain that the histogram doesn't reflect 14bit data when shooting raw? The inaccuracy of the histogram results from the additional headroom that results from 14bit data that is not accounted for in the supplied histogram,

I believe you are mistaken. The inaccuracy of the histogram results from the fact that it goes through many processing steps as part of creating the JPEG, each step altering the histogram. These alterations would have the same negative affects even if the camera and histogram were both the same bit depth.


My understanding is that the bit depth does not relate to the histogram
except that it limits the maximum resolution (precision of the x axis),
and even 8 bits is more than enough resolution for my purposes. That
is, after the maximum 14-bit value (16383) is mapped to the maximum
8-bit value in the histogram (255), that provides enough levels to
describe how bright something is for me.


White balance seems to be the biggest offender. A custom white balance under tungsten light, for example, can increase the blue channel by over two stops and decrease the red channel by one stop. So the histogram may show red is not clipping, when in fact it is clipped by one stop.

Other steps include the picture style (tone curve, white balance, contrast, saturation, colors, etc.) and color space conversion (aRGB, sRGB).

The only workaround I know of, for now, is to try and reverse as much of the processing as possible with in-camera settings and a special custom white balance:

http://forums.dpreview.com/forums/re...ssage=26905476

That ruins the utility of the preview, metadata, etc., but for me, getting an accurate histogram is worth the price of all that (so I can get the perfect exposure and maximum dynamic range).

along with the fact that the tone curve you use to "develop" the RAW data will be different than the one the camera uses to create the histogram. This is why the histogram changes when you change picture styles, even though picture styles don't affect RAW data (i.e., the camera is using a different tone curve to "develop" the histogram from the RAW data depending on the selected picture style).

Agreed.

A sensor has only one actual sensitivity. Put another way, each pixel on a sensor captures so many electrons during a capture. For any given shutter speed and aperture there is no way to make it capture more or fewer of them. Any ISO above the base ISO is simply the result of increasing the gain, amplifying the actual signal (and noise).

Agreed. Excellent description.

For the "L" and "H" settings, the increased sensitivity isn't even the result of an amplication. It is a basically a software manipulation. In the case of images shot in one of the H modes, they are basically ISO 6400 shots with a metadata notation (in the case of RAW files);

I *wished* that's how it worked!

Unfortunately, what they are actually doing is applying the digital push in the camera before the raw file is written. The Canon 10D was the only one that did it the right way, with metadata, as you describe. All the others since, including the 5d2, throw away the highlights for no reason.

The highest analog amplification in the 5D2 is 3200. (6400 is digital, even though it's not an "Expansion" ISO.) That means ISO 25,600 is a three stop digital push, with three stops of deleted headroom. The reason I said 5 stops is that I often prefer to use ISO 800 over 3200, because the read noise is not much higher in ISO 800, and I'm happy to trade a little read noise for *two* whole stops of additional head room (more total dynamic range, even though the noise floor is a little higher).

If Canon let the photographer choose to set analog ISO and metadata ISO separately, then one could use a metadata ISO of 25600 (analog 800), giving five stops more highlights than the current ISO 25600. That's why I said "Stop deleting up to 5 stops of highlights from the RAW data (i.e. allow ISO to be metadata)."

likewise, L shots are ISO 100 shots with a metadata notation.

You are correct about that. HTP is also similar to what I'm asking for, but it only allows one stop, and it's not enabled for all ISO settings.

no company is going to delete 5 stops of highlight data if there is an easy way to avoid doing so.

Yet they are. I think the reason is three-fold:

1. Customers accept all sorts of compromises at high ISO, even ones that are completely unnecessary (like this). Many photographers never even use high ISO.

2. Some customers and raw processing programs might get confused by the metadata. They break compatibility withevery new camera anyway, so I can't fathom why this would be an issue.

3. Canon had ISO metadata in the 10D, but removed it. Maybe there is a beaurocratic issue, such as edict from Managament/Marketing that is preventing the software engineers from doing the obviously correct thing.

If Canon was too stupid to take advantage of a way to squeeze five additional stops (!) out of their sensors, someone else would.

Other manufacturers are doing just that: MFDB and many digicams implement ISO as metadata. Of course, they have terrible high ISO read noise, so they don't even *need* the feature like Canon does.

* Stop clipping highlights in raw files taken at f/2.8 or less just to compensate for sensor AOR.

I'd be interested in a more detailed exposition of this point. I've never read this claim and I'm not sure what you mean by "sensor AOR".

I was being laconic; I meant "Angle of Response". Canon does a digital "push" at wide f/stops to compensate for lower sensor response to light from oblique angles (due to the microlenses).

The actual affect is pretty small (less than 1/3 stop), but it's the principle of the thing: leave the raw file alone. I haven't actually tested the 5d2 yet to see if it does this manipulation, but you can read about where it's done in the 30D here:

http://www.pages.drexel.edu/~par24/r...awScaling.html

* Stop wasting huge amounts of storage on ludicrous compression like sRAW.

I won't use this mode either but many folks will -- including alot of pro wedding photogs.

I highly desire a smaller file size and I would use it a lot if Canon built a *good* option. Instead, we have sRAW, with nasty aliasing artifacts and much lower resolution.

What is the purpose of sRAW? To reduce file size.

How does it achieve this? By throwing away resolution.

It's like a plane with 21 passengers and a cargo hold full of garbage. When the plane is too heavy, Canon immediately resorts to throwing passengers out of the plane, while keeping all the garbage.

That's why sRAW is the laughing stock of file formats. In the 50D, half of the sRAW2 filesize is wasted on two embedded jpegs, and the rest only contains a quarter of the resolution.

First of all, there's a simple nonlinear bit-depth compression they could do, like Nikon. The difference is invisible to the human eye, but that alone saves a tremendous amount of disk space.

Next, they could make the embedded jpegs optional, at the cost of making reviews a little slower. (I don't use the embedded jpegs anyway since Canon also doesn't bother providing truly raw RGB histograms).

After that they could compress bit-depth on a scale of exposure. You don't need 14 bits to record a scene at ISO 12800. Six or maybe 7 bits would be enough to encode the entire signal down to the noise floor. For every stop that exposure is decreased, the noise floor gets higher. The bit depth should be decreased to match the noise level. This compression wouldn't help or affect base ISO shots at all, but would make a huge difference for high ISO shots.

On the subject of bitdepth itself, the noise in the camera at base ISO isn't even clean enough to require 14 bits. 13 bits is more than enough to encode the entire range of signals as well as a half-stop of noise, and maybe even 12 would suffice.

Now that's four things that reduce filesize without *any* affect on the resulting image. Still has full resolution, RAW, etc. My calculations show that for a high ISO shot, file sizes would be at least 1/4th what they are now. That's 6 MB for the 5d2. I'd much rather have that then the half-resolution 15 MB files Canon has now.

But why stop there? There is an entire world of compression they could apply to the image with imperceptible affect on image quality. Advanced compression technology exists to reduce file size in ways that are visually lossless to the eye. At the very least, the "lossy" effect will be much smaller than just dumping entire pixels like sRAW. Unlike the methods I've already discussed, many of these methods, I'm sure, require some CPU power in the camera.

If the camera is too weak to do any sort of medium-to-advanced compression (and it probably is), then sRAW would be an acceptable compromise. But not until all the better ideas, the least of which are outlined above, which do not require any camera CPU power, are already implemented.

Most people will be VERY pleased with 30p. Of course, ideally you should be able to choose.

Agreed.

[tlinn]

I appreciate your taking the time to expand upon your orginal comments and respond to mine, Daniel. You make some good points. I appreciate the references as well. Looks like I have some reading to do this afternoon!

[dougthebiker]

Thanks for the feedback. I got to try a 5DII this afternoon with the 70-200L IS USM and had three of the neighborhood kids run straight toward me from about 50 yards away a few times.


Exactly 50% of the shots were OOPF!


Do you get better than that with a 40D or 50D? I thought the 5dII had the same autofocus as those two.