Is 1080P Broken On All Canon DSLRs?

[Viet Bach Bui]

3 hours ago, Inazuma said:

There has yet to be a Canon DSLR with a detailed 1080p image. Even Nikon is ahead of them now in this regard with the D500 and D5.

https://***URL removed***/reviews/image-comparison/fullscreen?attr29_0=canon_eos1dxii&attr29_1=panasonic_dmcgh4&attr29_2=nikon_d500&attr29_3=canon_eos80d&attr72_0=1080&attr72_1=1080&attr72_2=1080&attr72_3=1080&normalization=full&widget=378&x=-0.2454251883745963&y=0.16045765603694293

Wow the A6300 has the sharpest 4k image by far, even when compared to Sony's own A7 line. The 1080p is bad though.

[David Bowgett]

That's because the A6300 oversamples from a 6.5K area of the sensor. By comparison, the A7Sii is natively 4K, and the A7Rii's oversample is only something like 4.8K.

[Luke Mason]

1 hour ago, Justin Bacle said:

The problem with 3x3 binning is that every group of 9 pixels is different from its neighbours (because of the Bayer RGGB pattern). 2x2 binning would give great results though (as every group of 4 pixels would give true RGB value, like in the cinealta cameras i think), I don't know why they don't do it :s

It's called colour-aware pixel binning, which Canon used on C300 to produce 2K 444 image. It's different from simply reading out a group of pixels.

[tugela]

20 hours ago, DBounce said:

Had some spare time so I decided to look more closely into the 1080P of the 1DXMk2. And after doing this I concluded that it is pretty bad. More like 720P upscaled to 1080P.
Below I share my finding and compare to mobile as I firmly believe that a $6000 camera should handily outperform a smartphone if the playing field is even. The results may surprise you.

 

I think the problem is that most Canon DSLRs use 1080p of raw pixels, but after debeyering you end up with an effective color resolution of ~700-750p. Unless the sensor is oversampled you will not approach true 1080p resolution. My guess is that sampling is controlled through the hardware encoder in the processor, so if your camera still has one of the older processors (Digic 6 and earlier), you are going to get this soft relatively low res HD output. Maybe the Digic 7 will have the same problem, but it is too early to say since it is in only one camera for now.

Sampling is much better in 4K mode however (in those models that offer it), so in those cases you are probably better off shooting in 4K and downsampling to HD in post yourself.

6 hours ago, johnnymossville said:

I'm no Canon Fan Boi,  but what they lack in resolution they make up for in color science.     They have discovered the magic sauce that Sony would and SHOULD pay Billions for.  

There is no such thing as "color science". It is a made up buzz word (phrase) used in order to rationalize choices made.

[DBounce]

6 minutes ago, tugela said:

Sampling is much better in 4K mode however (in those models that offer it), so in those cases you are probably better off shooting in 4K and downsampling to HD in post yourself.

Yes,  but what if you would like to use the 120 FPS promised by Canon? That option is not available in 4k mode, max is 60 FPS.

[jcs]

19 minutes ago, tugela said:

There is no such thing as "color science". It is a made up buzz word (phrase) used in order to rationalize choices made.

What's a better, simpler term, etc., to describe a manufacturer's complete color processing system, from optics, color filter arrays, sensor design, and digital processing?

120fps is decent on the A7S II (perhaps A7R II also?). For the 1DX II 120fps, have you tried a little post-sharpening and perhaps unsharp mask with a radius around 30-150 pixels? (http://www.cambridgeincolour.com/tutorials/local-contrast-enhancement.htm). Add in decent film-like grain and it can be usable. Again, taking a still with the 1DX II as a reference and shooting with an A7x II and using the Canon still as a reference can result in decent color.

[Policar]

1 hour ago, tugela said:

I think the problem is that most Canon DSLRs use 1080p of raw pixels, but after debeyering you end up with an effective color resolution of ~700-750p. Unless the sensor is oversampled you will not approach true 1080p resolution. My guess is that sampling is controlled through the hardware encoder in the processor, so if your camera still has one of the older processors (Digic 6 and earlier), you are going to get this soft relatively low res HD output. Maybe the Digic 7 will have the same problem, but it is too early to say since it is in only one camera for now.

Sampling is much better in 4K mode however (in those models that offer it), so in those cases you are probably better off shooting in 4K and downsampling to HD in post yourself.

There is no such thing as "color science". It is a made up buzz word (phrase) used in order to rationalize choices made.

I'd never read that term until Red used it to justify their (at the time) bad colors. (Which are much nicer now.) Since it's totally subjective anyway just buy what looks good to you. 

If you look at a still JPEG from a Canon dSLR and compare it to a RAW image debayered in ACR the RAW file looks WAY sharper. Same difference you'd see when comparing h264 to RAW video. So basically, yeah.

Is the fine detail picture style available in video mode? I think it was introduced on the last few cameras. That has a better debarring algorithm for JPEGs at least. Still fairly limited DR, but whatever.

[manufilm]

7 hours ago, David Bowgett said:

The weird thing is that going by that link, the Canon DSLR that produces the best 1080p image actually seems to be the 5DS(r). Which fits with the rumours that its sensor produces a ~2.5K video signal due to essentially being a jumbo-sized 7D Mk II sensor, and they just downscale that signal to get 1080p.

From my experience with using Magic Lantern, and understanding of Canon cameras either do:

Canon 5D Mark IV/1Dc/1DX Mark II 4K:

1x1 (4K center crop)

Canon 5D Mark II/7D/T2i/70D and other 1080p Canon cameras released until 2014 (except the 5D Mark III):

3x3 (pixel binning horizontally, pixel skipping vertically, slight upscale) moire/aliasing/scaling artifacts

- Canon 5d Mark II: 1872x1053  effective upscaled to 1920x1080

- Canon 7D/T2i: 1728x972  effective upscaled to 1920x1080

- Canon 70D: 1824x1026  effective upscaled to 1920x1080

Canon 5D Mark III/IV/1DX Mark II/7D Mark II/80D 1080p:

3x3 pixel binning - 5d Mark III and most other Canon cameras since 2014 (better moire/aliasing and ISO performance than using pixel skipping)

- 5D Mark III (1920x1080 effective) sensor resolution chosen to avoid upscaling (5760/3=1920)

- 7D Mark II/1DX Mark II: 1824x1026 effective upscaled to 1920x1080

- 80D: 2000x1125 downscaled to 1080p (my guess)

- 5d Mark IV: 2240x1260 downscaled to 1080p (my guess)

Canon 720p Modes:

3x5 (pixel binning horizontally, pixel skipping vertically)

 

The Canon 5DS(r) is probably using 3x3 pixel binning:

2880x1620 resolution downscaled to 1080p (which happens to be the exact resolution of the Arri Alexa!)

 

[Luke Mason]

1 minute ago, manufilm said:

From my experience with using Magic Lantern, and understanding of Canon cameras either do:

Canon 5D Mark IV/1Dc/1DX Mark II 4K:

1x1 (4K center crop)

Canon 5D Mark II/7D/T2i/70D and other 1080p Canon cameras released until 2014 (except the 5D Mark III):

3x3 (pixel binning horizontally, pixel skipping vertically, slight upscale) moire/aliasing/scaling artifacts

Canon 5D Mark III/IV/1DX Mark II/7D Mark II/80D 1080p:

3x3 pixel binning - 5d Mark III and most other Canon cameras since 2014 (better moire/aliasing and ISO performance than using pixel skipping)

- 5D Mark III (1920x1080 effective) sensor resolution chosen to avoid upscaling (5760/3=1920)

- 7D Mark II/1DX Mark II: 1824x1026 effective upscaled to 1920x1080

- 80D: 2000x1125 downscaled to 1080p (my guess)

- 5d Mark IV: 2240x1260 downscaled to 1080p (my guess)

Canon 720p Modes:

3x5 (pixel binning horizontally, pixel skipping vertically)

 

The Canon 5DS(r) is probably using 3x3 pixel binning:

2880x1620 resolution downscaled to 1080p (which happens to be the exact resolution of the Arri Alexa!)

 

hahaha, you overestimated Canon, pre-5D III cameras use line skipping, there was no 3x3 binning at that time.

5D3 uses large group pixel bins of 5x5, resulting a 1152x648 RAW bayer, which then gets upscaled to 1080p, that's why it has about 600-700 lines of resolution

similar thing happens to other Canon cameras.

The best HD image from Canon DSLR is 1DC in S35 mode, which I've detailed in another post, it resolves near 1000 lines.

A 1824x1026 RAW bayer would be sharp as f***, with resolution easily exceeding 1000 lines.

[Policar]

21 minutes ago, Luke Mason said:

hahaha, you overestimated Canon, pre-5D III cameras use line skipping, there was no 3x3 binning at that time.

5D3 uses large group pixel bins of 5x5, resulting a 1152x648 RAW bayer, which then gets upscaled to 1080p, that's why it has about 600-700 lines of resolution

similar thing happens to other Canon cameras.

The best HD image from Canon DSLR is 1DC in S35 mode, which I've detailed in another post, it resolves near 1000 lines.

A 1824x1026 RAW bayer would be sharp as f***, with resolution easily exceeding 1000 lines.

Just curious, and not that I don't believe you, but what evidence do you have of this? 

The only thing that doesn't add up is that 1824X1026 would only resolve about 500 lines of resolution without aliasing according to nyquist, or 700-800 lines or so with efficient debayering algorithms that do allow for some aliasing (which are typical to most cameras). The 1000+ figure assumes a level of efficiency that's totally unprecedented. And that the 5D Mark III is producing a 1920X1080 RAW according to magic lantern, and its (very soft, let's be honest) image is no softer per-pixel than a JPEG straight from the camera. Upscaling would suffer from even softer per-pixel performance. Also, why make a sensor that's 3X HD only to bin to 3/5 of that? Just curious where you're getting the 5x5 figure because it doesn't correlate with real-world performance in any way.

The above figures (the post before yours) make a lot more sense to me. So does the mix of line skipping and pixel binning in different dimensions for the earlier generations correlate with ISO performance being about 1.5 stops worse in video than in JPEG before the 5D III. The result is still super soft and not exactly great... but we all know that.

I haven't done much research into this and it sounds like you have, but the figures posted before what you wrote correlate closely with camera performance and what you've quoted doesn't. Of course no one knows for sure except Canon engineers, but I suppose maybe someone at Magic Lantern has a pretty good idea, and I was curious if you were quoting someone from Magic Lantern or if you'd figured this out on your own and how.

[manufilm]

13 minutes ago, Luke Mason said:

hahaha, you overestimated Canon, pre-5D III cameras use line skipping, there was no 3x3 binning at that time.

5D3 uses large group pixel bins of 5x5, resulting a 1152x648 RAW bayer, which then gets upscaled to 1080p, that's why it has about 600-700 lines of resolution

similar thing happens to other Canon cameras.

The best HD image from Canon DSLR is 1DC in S35 mode, which I've detailed in another post, it resolves near 1000 lines.

A 1824x1026 RAW bayer would be sharp as f***, with resolution easily exceeding 1000 lines.

I did say the pre-5D3 use line skipping (pixel binning horizontally, pixel skipping vertically) This has been posted online and tested by others.

The 3x3 or 3x5 read modes are based on the options available by using Magic Lantern firmware.

[Luke Mason]

1 minute ago, manufilm said:

I did say the pre-5D3 use line skipping (pixel binning horizontally, pixel skipping vertically) This has been posted online and tested by others.

You said it's 3x3, suggesting both horizontal and vertical binning.

[jcs]

To get the best detail 1080p out of the 1DX II: use the Fine Detail picture profile with Contrast at -2. Skintones are still a bit waxy: use a little film grain to improve. Some aliasing will occur on fine detail, Neat Video might be able to help there. I tested this at 1080p 23.98 IPB and 119fps ALL-I on the 1DX II (needs a bit more post sharpening- still fairly soft even with Fine Detail picture style). Also tested the A7S II at 119fps- far more detail and usable DR, much lower noise, though does look more video-like vs. the Canon straight out of camera.

[Luke Mason]

26 minutes ago, jcs said:

To get the best detail 1080p out of the 1DX II: use the Fine Detail picture profile with Contrast at -2. Skintones are still a bit waxy: use a little film grain to improve. Some aliasing will occur on fine detail, Neat Video might be able to help there. I tested this at 1080p 23.98 IPB and 119fps ALL-I on the 1DX II (needs a bit more post sharpening- still fairly soft even with Fine Detail picture style). Also tested the A7S II at 119fps- far more detail and usable DR, much lower noise, though does look more video-like vs. the Canon straight out of camera.

Fine detail does not work in video mode, it returns to traditional sharpness setting with no control over sharpening radius and threshold.

[jcs]

45 minutes ago, Luke Mason said:

Fine detail does not work in video mode, it returns to traditional sharpness setting with no control over sharpening radius and threshold.

I tested using EOS Utility 3 and it allows changing Strength, Fineness, and Threshold. Fineness & Threshold changes didn't make any visible changes;  setting Sharpness Strength to just right of center on any picture style looks the same as the default Fine Detail.

[Luke Mason]

1 hour ago, jcs said:

I tested using EOS Utility 3 and it allows changing Strength, Fineness, and Threshold. Fineness & Threshold changes didn't make any visible changes;  setting Sharpness Strength to just right of center on any picture style looks the same as the default Fine Detail.

In video mode if you go to PP settings, fineness and threshold will be greyed out. Only strength is available, same effect as previous DSLRs

[jcs]

10 minutes ago, Luke Mason said:

In video mode if you go to PP settings, fineness and threshold will be greyed out. Only strength is available, same effect as previous DSLRs

Canon 1DX II, Video Mode, EOS Utility 3

All settings can be changed (did not see a visible difference for Fineness and Threshold while watching the Live View on the computer).

Canon 1DX II, Picture Style Settings for Fine Detail

Fineness and Threshold not present.

I haven't done detailed testing for in-camera vs. post-only sharpening (edge convolution, unsharp mask etc.), however the in-camera sharpening looked OK for this test. For slomo on the 1DX II I'd stick with 4K60p if quality is paramount, else 120 could be used for very short shots, and/or close ups along with Neat Video (aliasing, noise), post sharpening, and added film grain. If PDAF isn't needed, the A7S II is much better at 120fps and skin tones can match well enough using Canon reference stills (A7S II settings: Slog2 + SGamut3.cine, +12 sat. etc.). 

[Luke Mason]

1 hour ago, jcs said:

Canon 1DX II, Video Mode, EOS Utility 3

All settings can be changed (did not see a visible difference for Fineness and Threshold while watching the Live View on the computer).

Canon 1DX II, Picture Style Settings for Fine Detail

Fineness and Threshold not present.

I haven't done detailed testing for in-camera vs. post-only sharpening (edge convolution, unsharp mask etc.), however the in-camera sharpening looked OK for this test. For slomo on the 1DX II I'd stick with 4K60p if quality is paramount, else 120 could be used for very short shots, and/or close ups along with Neat Video (aliasing, noise), post sharpening, and added film grain. If PDAF isn't needed, the A7S II is much better at 120fps and skin tones can match well enough using Canon reference stills (A7S II settings: Slog2 + SGamut3.cine, +12 sat. etc.). 

I tried that a while ago, EOS utility seems to be a bug, it allows you to change the parameters but only strength affects the final image. It's the same video sharpening effect in all previous DSLRs, large radius with obvious edge halo. I just turn it off to 0.

[jcs]

33 minutes ago, Luke Mason said:

I tried that a while ago, EOS utility seems to be a bug, it allows you to change the parameters but only strength affects the final image. It's the same video sharpening effect in all previous DSLRs, large radius with obvious edge halo. I just turn it off to 0.

It's too bad there aren't any commercially available Super Resolution sharpening algorithms available (use aliasing information to actually improve resolution). That would be another solution.

 

When the 5D Mark III came out in 2012 everyone online complained about how soft it was. This video showed that with a little post sharpening and very sharp lenses it was usable: 

It's now 2016 and Canon still has the softest 1080p for DSLRs (except perhaps the 5DSR). It still takes some post work to make look decent, however the PDAF makes it almost a toss up between something like the A7S II which is very sharp but has almost useless AF. The C100 II is really the best bet for those looking for decent 1080p, good skintones, and PDAF for a decent price. For slomo Sony is currently the best bang for the buck.

 

[Luke Mason]

13 minutes ago, jcs said:

It's too bad there aren't any commercially available Super Resolution sharpening algorithms available (use aliasing information to actually improve resolution). That would be another solution.

 

When the 5D Mark III came out in 2012 everyone online complained about how soft it was. This video showed that with a little post sharpening and very sharp lenses it was usable: 

It's now 2016 and Canon still has the softest 1080p for DSLRs (except perhaps the 5DSR). It still takes some post work to make look decent, however the PDAF makes it almost a toss up between something like the A7S II which is very sharp but has almost useless AF. The C100 II is really the best bet for those looking for decent 1080p, good skintones, and PDAF for a decent price. For slomo Sony is currently the best bang for the buck.

 

5DS R uses a combination of line-skipping and pixel-binning to read its huge pixel count, the resulting strong aliasing made it appear sharp. But it's still problematic.