[Enc0der]

Who cares?
This pixel-peeping is getting even more ridiculous with video, eh?
You are recording and watching VIDEO.
Who cares about some purple pixels that are absolutely NOT VISIBLE in the video playback (in 100% size!) and can only be found if you capture one frame and then zoom in to a ridiculous 400% size?!

[NunoBarbosa]

Thanks Falconeye, for the link! it's impossible otherwise to find it in vimeo because it doen't have any tag.

That clips look good,
the fringe is visible, but comparing with regular videocameras, and the kind
of artifacts of other video DSLRs i must say the k-7 footage it's one of the cleanest out there. They choose really well the compression process.

What concerns me more is the lack of manual controls.
We get aperture...not bad, but shutter and ISO it's really important.
This camera gets very noisy quickly bacause of the auto ISO.
When it's dark we like a dark image, not a grainy lighten place.
It will be great to have that control to more serious work,
Pentax are you listening?

[thibs]

Folks from the German K-7 gathering in Hamburg have published an original, i.e., non recoded, video from the K-7. An internet first, it seems.

From the data of it, it is 720p, "**" quality (*** would be 58 MBit/s, ** 41 MBit/s, the video is 44 MBit/s).

The Codec is identified as:
Video: "Motion JPEG OpenDML" 30fps (not 29.97!)
Audio: "16 Bit Mono 32 kHz"

The video is here:

and you must log in to download the AVI original file (6s, 31 MByte).

(c) reisschuessel @ DFN/phototalk24.de phototalk24.eu :: Thema anzeigen - Bilder/Videos/Lob/Kritik K-7

Hint: "Nero Showtime" plays the MJPEG codec much more smoothly on slow computers than QT or Windows Mediaplayer.

Thank you much Falconeye. Quite nice IMO. At last better than the minimum required for me to be interested in the Video function of the K-7.

BTW, it means Nero is something acutally useful ?

[karl79]

Who cares?

It may not matter if you're out filming your family on holidays, but as a competitor to professional video industry standards, which video DSLRs are about to become, these issues are important and do make a difference.

[falconeye]

I am sorry if my post was taken to be meant to tear down the K-7 video function.

I did not mean so, the K-7 video function is great!

It is just about the point that I was interested if Pentax did anything to improve over the K20D burst mode and I found out that, besides increasing 21fps to 30fps and increasing 5s to 1800s, and adding audio, and wrapping the JPGs into MJPEG, they did not So, I thought I should share this bit of information as I know that others were watching, too.

Who cares about some purple pixels that are absolutely NOT VISIBLE in the video playback (in 100% size!)

I didn't care until I ran across this with my K20D video clips and found it disturbing. And it is visible at normal video playback. It "jumped to my eye" even when watching the smaller vimeo web playback.

I show the crops at 400% only because it makes it easier to identify the individual pixels, for the sake of easier discussion for those who care. And you need to identify pixels in any kind of Moiré discussion.

Also, some may find it noteworthy that K-7 720p is rescaled 768x1024 footage. Some may not, like the difference between 1080p and 720p doesn't show at normal viewing distances.

That clips look good,
the fringe is visible, but comparing with regular videocameras, and the kind
of artifacts of other video DSLRs i must say the k-7 footage it's one of the cleanest out there.

So, maybe this is a fair verdict then:

The K-7 video footage got some fringing artifacts but they are still less disturbing than video artifacts as known from other DSLR in its price range.

[raz]

falconeye I'm not sure if you can tell just by the purple fringing if the camera is using a set of superpixels or just 1 in 3 pixels to take a shot or not. This can be the result of downampling from a higher resolution depending of the algorithm used. My initial remarks are that the anti aliasing (that in K20D "video mode" is bad), in the k7 sample looks OK - and this mean it shouldn't look OK if the true video mode was only 1024x768, or maybe they improved the resizing algorithm ?. The only bad thing that I see is the frames aren't "crystal clear" how I expect from a 4000x3000 images resized to 1280x720

We need a direct mjpeg copy and a direct comparation (e.g.: shot a video of an object, then shot a photo of that object and resize the photo to 1280x720 to compare the sharpens,anti aliasing and details).

[falconeye]

From the bits of information gathered so far, I think I can summarize the video architecture now:

• The K-7 and K20D CMOS sensor has read-out channels capable to read out 25.6 million pixels/s. The K20D has 2 channels, the K-7 got 4.
• 2 channels are good for 3.5 fps full size or 21 fps 768x1024 video frames (21 = 6*3.5 (*)).
• 4 channels are good for 7.0 fps full size or 42 fps 768x1024 (every 6th pixel) or 21 fps 1536x1024 (every 3rd pixel) video frames.
• K-7 seems to feature shutter-limited 5.2 fps full size burst and no extra cost was involved to support the video mode as is!
• 1280x720p@30fps is resampled from 30 fps 768x1024 (768 px width!). As both resolution have a similar #pixels, this doesn't significantly degrade quality. The 1536x1024 mode won't look significantly better, though.
• The sub-sampling matrix is such that 1.5px-wide fringing artifacts at vertical high contrast borders can occur. K-7 and K20D share the same sub-sampling.
• The individual frames are up to ~300 kB large (K20D: 900 kB) which is quite a good quality (Photoshop 8-9/12 "high" ***). Due to the MJPEG codec, there are no additional time-compression artifacts or processor load.
• Because only every 6th pixel is read, low light performance doesn't benefit from supersampling and will look like noise when looked at at 100% crops. The effective sensor surface in video mode is 13.5 x 9 mm (sensitive surface to light, NOT a cropped surface or less DoF effect!!) or 1/0.6". This is still a lot bigger than P&S or camcorders (ranging from 1/3" to 1/1.5").

__
(*) As it appears, the read-out speed in reading every 6th pixel is exactly six times as fast as reading out the full sensor. Just as one would naively expect.

An emphasis on video would have meant that Pentax/Samsung would have increased the number of channels from 2 to 6, not 4. This is an interesting statement for the forthcoming Samsung NX too which is supposed to share the same sensor.


So, in summary:

The K-7 video outperforms camcorders for low light performance, compression artifacts, lens quality (at least along horizontal edges) and DoF without increasing the production cost of the camera (except maybe for the microphone).

---

Last edited by falconeye; 06-03-2009 at 02:17 PM.

[falconeye]

falconeye I'm not sure if you can tell just by the purple fringing if the camera is using a set of superpixels or just 1 in 3 pixels to take a shot or not.

This was a tricky part, indeed.

I needed to resample K20D video frames to 720p as well. This basically removes the bad aliasing artifacts, for K20D too. Not quite as good as restoring the 768x1024 internal frame first and resampling from there, but nearly so.

So, I compared the 720p K20D and K-7 frames side by side and it is the precise (equal) width of fringing which told me that it is coming from the same 2 pixel wide fringing on either side of an edge (in a 1536x1024 frame). Which again told me that the sub-sampling is the same.


Of course, there is a chance that I am wrong. So far, I am speaking from indirect evidences only. Like #channels as well. But they all say the same ...

[ecoronin]

this is the link to video samples soo cool http://www.pentaximaging.com/slr/K-7/

[NunoBarbosa]

this is the link to video samples soo cool PENTAX K-7 Digital SLR - Official PENTAX Imaging Web Site

I think they should have pic a more interesting subject...
apart from that, it looks it was made handheld and the SR works really good!
Defenitly a bonus over the competition...

[raz]

falconeye if you are right, I will be a little disappointed. Because as you mentioned it, noise will not be improved at low light. Having a ~4000x3000 6400 iso image resampled to 1280x720 is producing almost a free noise image. But by taking an image with only 1 in every 3 or 6 pixels, you have the same noise at the original 6400 iso image.

At least maybe they capture the frames at twice the size since now the sensor it hs double readout, e.g. let's say ~2000x1600 pixels and then resample them at 1280x720, but I don't think so, if they did so, they would almost certainly had a true 1080p mode

So although I don't want to agree with you, you may be right.

[nixcamic]

So, to sum everything up, the flaws are:

1. No 29.97 fps.
2. No 24 fps.
3. The rest of the world is getting screwed over by Europe's strange camcorder laws.
4. No manual.
5. Possible PF from the subsampling matrix.

It does shoot darn good quality video though, even with these problems. And great stills (which is its main purpose) , and I'd still buy one even if Pentax doesn't fix these problems (well depending on what the new samsung is like...)

[and]

oh finally the official sample. Well it looks good, artifact free.
I guess with the discovered issues if hte video feature becomes popular they will improve it in the next round

[and]

So, to sum everything up, the flaws are:

1. No 29.97 fps.

I didnt get that, why was that a problem? 30 fps instead of 29.97? is it because of audio sync

[nixcamic]

I didnt get that, why was that a problem? 30 fps instead of 29.97? is it because of audio sync

Yeah, and lining it up with other footage. You can just reconform everything, but its a pain.