Sony a77 appears to fake video specs!

Started Feb 26, 2013 | Discussions
aim120 Regular Member • Posts: 264
Re: Sony a77 appears to fake video specs!

Compare a still photo from any dslr to video grab and the still photo will always win in sharpness.

Also when you review a still image  taken from a SLT  on the EVF ,its looks much better then the Live view of the same image shown in EVF.

Briarios
Briarios Regular Member • Posts: 202
Re: Sony a77 appears to fake video specs!

aim120 wrote:

Compare a still photo from any dslr to video grab and the still photo will always win in sharpness.

Also when you review a still image taken from a SLT on the EVF ,its looks much better then the Live view of the same image shown in EVF.

Yes, agreed.

Not following the process-so-far of the guy from the link that the OP presented, though I understand that is what OP is trying to learn, what weight can be put on that test. Current thinking is, not a lot, but I'll wait for those more brillianter than I, to make that call

Comes to mind that as an exercise one can get full still video resolution video output from a57 or other DSLR's, but frame rate would be limited by the max. burst rate - simply by stitching together a stills burst into a movie file with mpeg streamclip.

 Briarios's gear list:Briarios's gear list
Casio Exilim EX-FH20 Sony Cyber-shot DSC-S90 Pentax Optio WG-2 GPS Sony SLT-A57 +1 more
IanML Senior Member • Posts: 1,653
Re: Sony a77 appears to fake video specs!

Briarios wrote:

I must be missing something. It appears that the OP in the link, is direct comparing a 6000x4000 still with a 1920x1080 video frame

You are.  He clearly stated that he downscaled the still and used that to compare to the single video frame.

I think the point is that the process he used for the downscaling is better than that used by the video system, and so it should be, as it is not constrained by processor capacity or time.

-- hide signature --

Ian

 IanML's gear list:IanML's gear list
Sony SLT-A33 Sony Alpha NEX-5N Sony SLT-A57 Sony a77 II +1 more
Shield3 Senior Member • Posts: 2,440
Re: Sony a77 appears to fake video specs!

TrojMacReady wrote:

Shield3 wrote:

Alphoid wrote:

The Sony a77 is advertised to shoot 1080p video. In fact, it appears the sensor shoots something around 720p, which is further cropped by 20%. Then the whole low-res video is digitally upsampled to 1080p.

http://www.mitros.org/p/a77/a77_video.html

Cute, huh? I wish I could at least get out a native resolution video stream. It would save size, and limit degradation from upsampling to non-native resolution.

What do folks think? I'll cross-post to other forums later.

You mention the 1.2x crop factor for the a77 - it's more like 1.36. The a99 with IS enabled is 1.2x. The net effect with the APS-C crop is your video on the a77 is 1.86x.

Actually, that would imply just over 1.2x (based on APS-C cropfactor of about 1.53x).

I felt the A77 cropped much more significantly than the a99 did.  Perhaps it's the illusion of APS-C; but a 10mm lens will be 18.6mm in video mode on the a77.

 Shield3's gear list:Shield3's gear list
Canon EOS 5D Mark III Canon EF 35mm F1.4L USM Canon EF 85mm f/1.2L II USM Canon EF 135mm F2L USM Canon EF 70-200mm f/2.8L IS II USM +2 more
Briarios
Briarios Regular Member • Posts: 202
Re: Sony a77 appears to fake video specs!

IanML wrote:

Briarios wrote:

I must be missing something. It appears that the OP in the link, is direct comparing a 6000x4000 still with a 1920x1080 video frame

You are. He clearly stated that he downscaled the still and used that to compare to the single video frame.

I think the point is that the process he used for the downscaling is better than that used by the video system, and so it should be, as it is not constrained by processor capacity or time.

-- hide signature --

Ian

Thanks. Agreed that EVEN IF the a77 video system were downscaling 1080P to 720P (doubtful), they wouldn't be comparable results.

 Briarios's gear list:Briarios's gear list
Casio Exilim EX-FH20 Sony Cyber-shot DSC-S90 Pentax Optio WG-2 GPS Sony SLT-A57 +1 more
Draek
Draek Senior Member • Posts: 2,028
Re: Sony a77 appears to fake video specs!

Amazing how you "feel" down to 0.1mm differences in focal length, when most people have trouble telling a 48mm apart from a 52mm. Still, "crop factors" are multiplied, not added, so as Troj said, if it were 1.86x total then the crop inflicted by video alone would be 1.21x, since 1.21 * 1.53 = 1.86.

 Draek's gear list:Draek's gear list
Fujifilm FinePix F70EXR Samsung TL500 Canon PowerShot A1200 Sony Alpha DSLR-A390
Spillicus
Spillicus Contributing Member • Posts: 678
Re: Sony a77 appears to fake video specs!

IanML wrote:

Briarios wrote:

I must be missing something. It appears that the OP in the link, is direct comparing a 6000x4000 still with a 1920x1080 video frame

You are. He clearly stated that he downscaled the still and used that to compare to the single video frame.

I think the point is that the process he used for the downscaling is better than that used by the video system, and so it should be, as it is not constrained by processor capacity or time.

-- hide signature --

Ian

I think you're right in that part of it is in the downscaling and sharpening done on the computer is better than on the fly.

The other part is that the A77 isn't outputing uncompressed video, it's all in H.264 in the AVCHD file.  Videos are not simply a collection of still images.  Even using some kind of a key frame involves reconstruction of the image, so it's not 100% the same as dealing with the still.

 Spillicus's gear list:Spillicus's gear list
Sony SLT-A65 Sony DT 18-55mm F3.5-5.6 SAM Sony DT 35mm F1.8 SAM Tamron SP 70-300mm F/4-5.6 Di USD
Shield3 Senior Member • Posts: 2,440
Re: Sony a77 appears to fake video specs!

Draek wrote:

Amazing how you "feel" down to 0.1mm differences in focal length, when most people have trouble telling a 48mm apart from a 52mm. Still, "crop factors" are multiplied, not added, so as Troj said, if it were 1.86x total then the crop inflicted by video alone would be 1.21x, since 1.21 * 1.53 = 1.86.

Don't be so snarky.  I'm saying when I threw a 24-70 on the a77 and switching from stills to video mode seemed quite a bit more of a crop than the same procedure on the a99.

 Shield3's gear list:Shield3's gear list
Canon EOS 5D Mark III Canon EF 35mm F1.4L USM Canon EF 85mm f/1.2L II USM Canon EF 135mm F2L USM Canon EF 70-200mm f/2.8L IS II USM +2 more
OP Alphoid Veteran Member • Posts: 5,333
Still not line skipping

An optical low-pass filter is the way to battle aliasing. The a77 does not have an OLPF (or it does not have one that would have any effect on a 2MP video; the AA filter is designed for a 24MP image).

A software low-pass filter would not solve moire. A low-pass filter, to reduce moire, has to come before the aliasing. After the aliasing comes into play, high frequency are already translated into low frequencies, and a low-pass filter will not effect low frequencies.

Regarding specific documents:

  • Good software downsampling, as in the F65 will eliminate moire without the need for an OLPF. Bad downsampling will give moire. Both will have identical or near-identical sharpness. 
  • The 5D explanation does not make sense for the a77. The a77 crops, so it is not tied to downsampling for misaligned resolutions. 
  • The D800 explanation, again, leads to moire and not softness. 

The original page has been updated with images with simulated binning and line skipping. In both cases, the output looks much better than the actual a77 output.

Just Having Fun Veteran Member • Posts: 3,869
Got it.

I always call that focus-assist and it is very helpful.  I am not sure why Sony crippled so much on these cameras.  The A65 is even worse.  You are limited to ISO 1600.

TrojMacReady
TrojMacReady Veteran Member • Posts: 8,725
Re: Still not line skipping

Alphoid wrote:

An optical low-pass filter is the way to battle aliasing. The a77 does not have an OLPF (or it does not have one that would have any effect on a 2MP video; the AA filter is designed for a 24MP image).

A software low-pass filter would not solve moire. A low-pass filter, to reduce moire, has to come before the aliasing. After the aliasing comes into play, high frequency are already translated into low frequencies, and a low-pass filter will not effect low frequencies.

There are many ways of sampling (software) that can influence aliasing. That's one of the reasons why so many sampling algorithms yield different results regarding aliasing (and moire).

Regarding specific documents:

  • Good software downsampling, as in the F65 will eliminate moire without the need for an OLPF. Bad downsampling will give moire. Both will have identical or near-identical sharpness.
  • The 5D explanation does not make sense for the a77. The a77 crops, so it is not tied to downsampling for misaligned resolutions.
  • The D800 explanation, again, leads to moire and not softness.

The original page has been updated with images with simulated binning and line skipping. In both cases, the output looks much better than the actual a77 output.

The Sony documentations shows there is a difference in sharpness too. And again, we're not talking about downsampling vs downsampling in your example, but downsampling vs subsampling. The latter comparison can definitely give subsantial differences.

The A77 crop is for IS purposes, just like with the A99 (which does allow to remove it when turning off IS). Not so much to change sampling resolution.

Debayering results in a loss of roughly 30% resolution per axis. Not so when you're downsampling from a much higher resolution as done in the F65, or with your still image from 24MP.

OP Alphoid Veteran Member • Posts: 5,333
Re: Still not line skipping

TrojMacReady wrote:

Alphoid wrote:

An optical low-pass filter is the way to battle aliasing. The a77 does not have an OLPF (or it does not have one that would have any effect on a 2MP video; the AA filter is designed for a 24MP image).

A software low-pass filter would not solve moire. A low-pass filter, to reduce moire, has to come before the aliasing. After the aliasing comes into play, high frequency are already translated into low frequencies, and a low-pass filter will not effect low frequencies.

There are many ways of sampling (software) that can influence aliasing. That's one of the reasons why so many sampling algorithms yield different results regarding aliasing (and moire).

Correct. This contradicts nothing I said.

The Sony documentations shows there is a difference in sharpness too. And again, we're not talking about downsampling vs downsampling in your example, but downsampling vs subsampling. The latter comparison can definitely give subsantial differences.

Your definition is incorrect.

Subsampling is one form of downsampling. In most applications, downsampling is implemented as a lowpass filter followed by subsampling. Subsampling is the special case where that lowpass filter is 1. Subsampling loses no resolution -- it just results in very bad aliasing. The page includes an example image where downsampling is implemented as subsampling (as a link in the bullet list at the bottom). You can compare. The image looks very good compared to the video which comes out of the a77.

(Footnote: The definition I give is not universal. In some fields, subsampling and downsampling mean the same thing. There is, however, no definition I am aware of which makes your statement correct.)

Beach Bum Senior Member • Posts: 1,055
Completely unscientific
5
 

Alphoid wrote:

The Sony a77 is advertised to shoot 1080p video. In fact, it appears the sensor shoots something around 720p, which is further cropped by 20%. Then the whole low-res video is digitally upsampled to 1080p.

http://www.mitros.org/p/a77/a77_video.html

Cute, huh? I wish I could at least get out a native resolution video stream. It would save size, and limit degradation from upsampling to non-native resolution.

What do folks think? I'll cross-post to other forums later.

There are many possible reasons for video that appears soft. To jump to the conclusion that it's 720p upsampled to 1080p is ridiculous.

And the idea of using still images as a surrogate for a video frame is no less ridiculous. Still images, even at the same resolution as a video frame (and from the same camera), often look crisper and sharper than a video frame. They also have entirely different color spaces and compression techniques, both favoring the still image over the video frame.

All in all, very weak "proof" and nowhere near a smoking gun.

TrojMacReady
TrojMacReady Veteran Member • Posts: 8,725
Re: Still not line skipping

Alphoid wrote:

TrojMacReady wrote:

Alphoid wrote:

An optical low-pass filter is the way to battle aliasing. The a77 does not have an OLPF (or it does not have one that would have any effect on a 2MP video; the AA filter is designed for a 24MP image).

A software low-pass filter would not solve moire. A low-pass filter, to reduce moire, has to come before the aliasing. After the aliasing comes into play, high frequency are already translated into low frequencies, and a low-pass filter will not effect low frequencies.

There are many ways of sampling (software) that can influence aliasing. That's one of the reasons why so many sampling algorithms yield different results regarding aliasing (and moire).

Correct. This contradicts nothing I said.

Actually it does. Most current jpeg engines do exactly that as well, remove moire and aliasing after the fact. And the Sony engine is one of the most effective in removing moire (turns out it does have a strength afterall... ).

The Sony documentations shows there is a difference in sharpness too. And again, we're not talking about downsampling vs downsampling in your example, but downsampling vs subsampling. The latter comparison can definitely give subsantial differences.

Your definition is incorrect.

Subsampling is one form of downsampling. In most applications, downsampling is implemented as a lowpass filter followed by subsampling. Subsampling is the special case where that lowpass filter is 1. Subsampling loses no resolution -- it just results in very bad aliasing. The page includes an example image where downsampling is implemented as subsampling (as a link in the bullet list at the bottom). You can compare. The image looks very good compared to the video which comes out of the a77.

(Footnote: The definition I give is not universal. In some fields, subsampling and downsampling mean the same thing. There is, however, no definition I am aware of which makes your statement correct.)

Semantics or not, the point stands that sampling by ignoring surrounding pixels when replacing them, is far more detremantal to the quality of output including resolution. Aliasing and moire (false detail as such) shouldn't be confused with real resolution to begin with.

atclen
atclen Regular Member • Posts: 151
Re: Completely unscientific

I'm with Beach Bum. We see video's all over You tube. The a77 can hold its on. I've seen a lot sharper than that sample used. Just the wrong f-stop could have cause that. Unscientific !!

 atclen's gear list:atclen's gear list
Samsung NX1 Samsung NX500 Samsung NX 30mm F2 Pancake Samsung NX 18-55mm F3.5-5.6 OIS Samsung NX 20mm F2.8 Pancake +12 more
OP Alphoid Veteran Member • Posts: 5,333
Re: Still not line skipping

TrojMacReady wrote:

Semantics or not, the point stands that sampling by ignoring surrounding pixels when replacing them, is far more detremantal to the quality of output including resolution. Aliasing and moire (false detail as such) shouldn't be confused with real resolution to begin with.

This is completely false and incorrect. Subsampling maintains all of the resolution possible in the original image up to the Nyquist frequency of the new image. It maintain as much detail as is possible to maintain at the new resolution. Algorithms can do worse for resolution than subsampling, but they cannot do better.

The problems with subsampling have to do with aliasing (and, in particular, aliasing of noise). In most domains, downsampling is done by first including some form of linear low-pass filter. This reduces high-frequency detail (including noise), so it cannot alias into the new image. This low-pass filter is rarely perfect, and so it often has the unfortunate side-effect of reducing resolution in the new image as well. This is what the AA filter in your camera does. Sensors without AA filters are closer to subsampling, and therefore maintain better resolution at the cost of aliasing.

In image processing, for performance reasons, instead of a linear filter, you often use some kind of non-linear hack (e.g. various spline fits), but for the purposes of this discussion, it is the moral equivalent.

Rab G Regular Member • Posts: 398
Re:I think this is far more interesting......
4

Hi All

I had a look at the link given in the op http://www.mitros.org/p/a77/a77_video.html

at the bottom of the link the following bullet point.

"This is the second time I've caught Sony cheating -- this time in technology; last time in terms of service".

Obviously this person has some kind of grudge against Sony so I would take his testing method which looks flawed and the results with a very large pinch of salt!!!!!

RabG

Briarios
Briarios Regular Member • Posts: 202
Re: Sony a77 appears to fake video specs!

Subject of test, is supported by folds of comforter.  Couldn't be expected to be stable over the period of test. No indication here that the frame chosen by the OP was a reference or key frame, or just an i frame.  OP states "...a77 unusable for video - logical fallacy.

The real question here is, why can't the OP get good crops of his a57 video?  I cropped a57 video at 40% last night and the results looked fine to me, comparative to the original - that is a rather extreme crop.

Anybody have experience cropping a77 video?

 Briarios's gear list:Briarios's gear list
Casio Exilim EX-FH20 Sony Cyber-shot DSC-S90 Pentax Optio WG-2 GPS Sony SLT-A57 +1 more
Briarios
Briarios Regular Member • Posts: 202
Re: Sony a77 appears to fake video specs!

Briarios wrote:

The real question here is, why can't the OP get good crops of his a77 video?

 Briarios's gear list:Briarios's gear list
Casio Exilim EX-FH20 Sony Cyber-shot DSC-S90 Pentax Optio WG-2 GPS Sony SLT-A57 +1 more
photo chris Regular Member • Posts: 281
Re: Sony a77 appears to fake video specs!
1

Alphoid wrote:

The Sony a77 is advertised to shoot 1080p video. In fact, it appears the sensor shoots something around 720p, which is further cropped by 20%. Then the whole low-res video is digitally upsampled to 1080p.

http://www.mitros.org/p/a77/a77_video.html

Cute, huh? I wish I could at least get out a native resolution video stream. It would save size, and limit degradation from upsampling to non-native resolution.

What do folks think? I'll cross-post to other forums later.

DSLR's only resolve somewhere around 700 lines, so none of them are true 1080p, compared to something like the EX1 which easily resolves more than 1000 lines.  My phone also claims to shoot 1080p HD video, while there may be 1080 vertical pixels, it certainly can't come close to the EX1's resolution, same goes for the Gopro and your A77.  There are plenty of resolution chart tests demonstrating this floating around the web.

Keyboard shortcuts:
FForum MMy threads