Canon 5D Mark III - Resolution Review

[psolberg]

peederj- I too have found the bitrates to be a bit low: detail is being lost by overcompression.

I did a detailed test of I-only vs. IPB. I-only is noisy, and appears to have a macroblock issue in PPro (couldn't replicate in FCPX on the Mac, though their sharpen function is Unsharp Mask looking and can't really bring out fine detail since there is no 'radius' option). I could not see any resolution/detail increase by using I-only; still recommend IPB until the issue(s) are worked out.

JasonATL- I have found 'Faithful' works the best so far. You can also sharpen fine detail with an Unsharp Mask (USM) filter with radius set to < 1.0. Local Contrast Enhancement via USM can be achieved with a 30-100 radius.

I don't understand how the bitrates can be too low. it is TWICE the bitrate of the 5DII and nearly 3X that of the IPB codec which IIRC is 28mbps. I've seen more detailed footage from nikon's internal IBP codec than from the all-I canon codec. I think the problem is sensor sampling. I suspect the way canon was able to reduce the amount of moire was by trading fine detail. so no matter how much bits per second you throw a it, a soft image is being created early enough before it is even compressed.

[JasonATL]

I've made a resolution and sharpness test of the 5D3 using a resolution test chart. I'm no pro, so I don't make more of this than it is. Still, I thought it performed quite well. My Canon 600D shows about 700 lines of resultion and this appears to show 800 or more lines. My Sony PMW-EX1 shows the full 1000 lines on the same test, which makes me think I've got it set up close to correctly - but I'm open to pointers and insights.

The video should be live on Vimeo in a few minutes.

https://vimeo.com/39517721

[peederj]

The bitrates they are advertising are maximum bitrates. In practice I'm getting something close to 60% of the advertised figures on steady-state shots. High-motion shots may get up to those bitrates. It's VBR, and a very stingy one at that. I want a superfine IPB mode with double what we're currently getting, and given ALL-I is rated as streaming 90mbps there's no way they can claim they can't give it to me if they are willing.

crippleware... 

[psolberg]

interesting post at EOSHD

http://www.eoshd.com/content/7727/james-miller-removes-optical-low-pass-filter-from-5d-mark-iii-for-resolution-increase

apparently, the soft video output of hte 5DmkIII may be due (at least in part) to the sensor OLP filter. This means anybody hoping that either a bitrate increase or a firmware update would result in more resolutions needs to keep their expecations in check. The problem may very well be hardware and if the data being sample is already smudged, then the image quality is as good as it is going to get.

[peederj]

interesting post at EOSHD

http://www.eoshd.com/content/7727/james-miller-removes-optical-low-pass-filter-from-5d-mark-iii-for-resolution-increase

apparently, the soft video output of hte 5DmkIII may be due (at least in part) to the sensor OLP filter. This means anybody hoping that either a bitrate increase or a firmware update would result in more resolutions needs to keep their expecations in check. The problem may very well be hardware and if the data being sample is already smudged, then the image quality is as good as it is going to get.

Easily disproven: take a look at the resolution 10x live view ("focus assist") gives you vs. what streams out of the cripple codec. The hardware is more than fine, and can generate amazing resolution.

[psolberg]

interesting post at EOSHD

http://www.eoshd.com/content/7727/james-miller-removes-optical-low-pass-filter-from-5d-mark-iii-for-resolution-increase

apparently, the soft video output of hte 5DmkIII may be due (at least in part) to the sensor OLP filter. This means anybody hoping that either a bitrate increase or a firmware update would result in more resolutions needs to keep their expecations in check. The problem may very well be hardware and if the data being sample is already smudged, then the image quality is as good as it is going to get.

Easily disproven: take a look at the resolution 10x live view ("focus assist") gives you vs. what streams out of the cripple codec. The hardware is more than fine, and can generate amazing resolution.

I don't know that I agree with your assesment but removing the OLP filter seems to help. I'm sure more testing will be done to prove or disprove that modification as a way to fully realize the potential of the camera. I guess stay tuned.

[JasonATL]

I'm with peederj on this on. This is simply illogical. If the optical low pass filter is, indeed, optical, then it is in hardware. In live view, zoom in as peederj suggests. The resolution is there on the sensor AFTER the optics of the camera. So, any softening would logically come from the downconversion AFTER the sensor and the optics.

Having said that, if you provide an even sharper (dare I say even aliased) image to the same downscaling algorithm, it seems plausible to me that the result might appear sharper. But, that won't necessarily make a better image overall.

I won't be running out to mod my 5D3 any time soon, assuming this is even serious.

[psolberg]

I'm with peederj on this on. This is simply illogical. If the optical low pass filter is, indeed, optical, then it is in hardware. In live view, zoom in as peederj suggests. The resolution is there on the sensor AFTER the optics of the camera. So, any softening would logically come from the downconversion AFTER the sensor and the optics.

Having said that, if you provide an even sharper (dare I say even aliased) image to the same downscaling algorithm, it seems plausible to me that the result might appear sharper. But, that won't necessarily make a better image overall.

I won't be running out to mod my 5D3 any time soon, assuming this is even serious.

just curious, did you watch the video? phillip bloom had his own take
http://philipbloom.net/2012/04/01/a-drastic-solution-to-increasing-sharpness-with-the-5dmkiii/

To my eyes, it looks better, but as I said, I'm sure we'll find out more in time.

[jcs]

I compared P.Bloom's sharpened material to J.Miller's OLPF hack on a computer monitor and on an HDTV (D8000). Though it's not the same subject matter, I don't see any significant advantage to no-OLPF vs. post sharpening*. I did see aliased edges with the no-OLPF footage, however.

*There is at least one disadvantage to post sharpening- it effects edges of bokeh. Fortunately, sharpening can be controlled, per shot. When noise is added as 'film grain' (not straight Gaussian noise- processed a bit), it helps fix issues caused by sharpening in post: https://vimeo.com/39523633 (link to instructions on dvxuser).

[psolberg]

I compared P.Bloom's sharpened material to J.Miller's OLPF hack on a computer monitor and on an HDTV (D8000). Though it's not the same subject matter, I don't see any significant advantage to no-OLPF vs. post sharpening*. I did see aliased edges with the no-OLPF footage, however.

*There is at least one disadvantage to post sharpening- it effects edges of bokeh. Fortunately, sharpening can be controlled, per shot. When noise is added as 'film grain' (not straight Gaussian noise- processed a bit), it helps fix issues caused by sharpening in post: https://vimeo.com/39523633 (link to instructions on dvxuser).

yes. sharpening just to achive what IMO should be the output we should get unedited blows big time. Not only do you have to be selective which ads work in post but you have to watch out the amount or introduce artifacts. I know there are ways to improve the footage we've seen so far but it is just so annoying to have to add it each and every time......

[LetTheRightLensIn]

I'm with peederj on this on. This is simply illogical. If the optical low pass filter is, indeed, optical, then it is in hardware. In live view, zoom in as peederj suggests. The resolution is there on the sensor AFTER the optics of the camera. So, any softening would logically come from the downconversion AFTER the sensor and the optics.

Having said that, if you provide an even sharper (dare I say even aliased) image to the same downscaling algorithm, it seems plausible to me that the result might appear sharper. But, that won't necessarily make a better image overall.

I won't be running out to mod my 5D3 any time soon, assuming this is even serious.

just curious, did you watch the video? phillip bloom had his own take
http://philipbloom.net/2012/04/01/a-drastic-solution-to-increasing-sharpness-with-the-5dmkiii/

To my eyes, it looks better, but as I said, I'm sure we'll find out more in time.

It still doesn't look at all as sharp as the C300 though. It's hard to see how it could help more than a modest little bit. I mean take an unmodded 5D3 22MP still and downsample to 2MP and it looks way sharper than the video a stock 5D3 produces AND simply look at a 100% view of a 22MP still and it still looks way sharper, so if the AA is so strong as to blur up 2MP off the sensor then how can a 22MP look at the sensor look so much vastly crisper?

I still think they need to add a 2x2 1.6x cropped true 1920x1080p mode on the 5D3. A crop mode would be useful at times anyway. Nikon and the others offer cropped modes. But Canon and their stupid protectionism are shooting themselves in teh foot. They could've had raves beyond raves for this had they just added zebra strips, zoom focusing while shooting, a true 1920x1080 mode even if it perhaps could only happen at 1.6x crop due to technical limitations (3x3 may be too large block for the internal AA filter to help enough so maybe they had to blend things, or perhaps, as some suggest 3x3 doesn't work well and they are doing 4x4 but at reduced resolution then).

I wonder if they didn't cripple the 5D3 to save the Cinema line.... and yet it was the 5 series video were they actually were the revolution, with their new cinema line they will just be one among many.

[psolberg]

It still doesn't look at all as sharp as the C300 though. It's hard to see how it could help more than a modest little bit. I mean take an unmodded 5D3 22MP still and downsample to 2MP and it looks way sharper than the video a stock 5D3 produces AND simply look at a 100% view of a 22MP still and it still looks way sharper, so if the AA is so strong as to blur up 2MP off the sensor then how can a 22MP look at the sensor look so much vastly crisper?

I don't know the details of the sampling canon does for video versus stills, but the quality without the OLP is undeniably better which means that at least part of the issue is hardware related and not firmware. If the OLP filter played no part in the final output, then the video footage would look nearly identical with the exception of slightly more aliasing. However I'm seeing a noticeable increase in detail and it has yet to be explained away by those that don't think the OLP makes any difference. I agree it is no C300, but that's besides the point, it is better than the stock 5DIII and that's our reference point. Until I see otherwise, I can only believe what I see, and what I see is a noticeable impact of the OLP sensor based on empirical evidence.

[peederj]

Has the guy who removed the OLPF posted the absolute most dead obvious first step of a resolution chart shot with modded and unmodded bodies with downloadable direct from camera files and enough information to 100% reproduce the results? If not, he's hiding something, and I think his inducing people to damage their $3500 cameras is seriously outrageous if he is.

[JasonATL]

Has the guy who removed the OLPF posted the absolute most dead obvious first step of a resolution chart shot with modded and unmodded bodies with downloadable direct from camera files and enough information to 100% reproduce the results? If not, he's hiding something, and I think his inducing people to damage their $3500 cameras is seriously outrageous if he is.

I agree with the idea that a resolution chart (comparison) would be useful here. As much as I like resolution charts, it is picture quality in the real world that matters most. A resolution chart would tell us if resolution is really improved, but it wouldn't necessarily answer how much aliasing shows up on certain objects (though it should show some aliasing, if it exists).

I don't think James Miller is really trying to pull one over on people, nor is he necessarily encouraging anyone to damage their own cameras. Anyone who does so based only on his two videos might deserve a $3500 paper weight.

What he has indicated will be forthcoming this week is a side-by-side comparison with an unmodded 5D3 and his modded 5D3.

[JasonATL]

I'm hoping that someone more knowledgeable than myself can help us interpret this.

Here is capture from the same source video as in the resolution chart tests that I did (video linked above in this thread). In this capture, the video is on the bottom and the still (scaled to 1920x1080p) is on the top. I applied an EXTREME amount of sharpening to the video (more than anyone in their right mind would even dream of applying to real shots) and some to the still.

Under the theory (as I understand it) that sharpening cannot create resolution, I wonder what is going on in this chart? There are distinguishable groups of lines near the "10" (1000 lines) mark.

Does this mean that the 5D3 might have more true resolution than we think?

I don't think the actual resolution makes or breaks this camera. Though I want to get more real-world footage, I'm happy with what I'm seeing so far. I'm downright amazed at how much sharpening I can apply to real shots and it still look very nice. I'm just curious about this and had already shot the charts (a bit of a painstaking process - and it probably still wasn't done perfectly).

I'm just hoping someone can improve my understanding of how to read and interpret what we see in these charts.

Thanks!