Forgot Password
Pentax Camera Forums Home
 

Reply
Show Printable Version Search this Thread
06-03-2009, 06:50 AM   #556
Senior Member




Join Date: May 2009
Location: Portugal
Posts: 112
1536 x 1024 sample



log in vimeo to download raw sample

06-03-2009, 06:52 AM   #557
Veteran Member
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,867
Original Poster
K-7 video frame FINAL analysis

I am now able to present my final analysis of the K-7 video engine.

At , there now is an original 1536x1024 video file from the K-7.

According to JCPentax, this is the "native" or "full" resolution all video sizes are downsampled from. So, this is the best size to base an analysis on.

In the attachment, you find additional material backing up my findings.

In summary: My earlier verdict, which contained speculative elements, is now fully confirmed. IMHO, nothing is left in the dark anymore.

Details:

The crops at 1536 x 1024 show a good resolution along an axis in vertical direction but along a horizontal direction, pixels are basically duplicated (well, no exactly, but almost). This has been exactly the same for K20D burst mode images.

This means that the internal video capture is at 768x1024 pixels (2:1 pixel aspect ratio!), or rotated SVGA resolution. I already antipated this when drawing my subsampling matrix above. Please, don't quote this resolution as 1024x768!

JCPentax may claim it is 1536x1024 internally. But then this is just what engineers told their marketing. They and now I know better. The doubled vertical edge borders and the resulting fringing (and its size and color) speak a clear language. There actually is a bit of non-redundant information between odd and even pixel columns. So, the claim of 1536x1024 being the internal resolution isn't exactly wrong, it is just that it is misleading.

My personal guess is that the 6x3 subsampling matrix is used twice to construct an RGB pixel, yielding two sets of 768x1024 pixels, however with very similiar data. Both sets are then interlaced for one (jaggy) 1536x1024 frame.

And all this leads to rather ugly aliasing and fringing artifacts when watching the movie. This is not to be ignored anymore and not a "pixel peeper's fantasy". Just download and watch the movie at full size.
Disclaimer: At 720p, the videos look nice. Here, I am talking about the 1536x1024 mode.

For the K20D, I had published a method to remove these artifacts for 1080p footage. You find it crosslinked in my earlier analysis within this thread. The 2nd attachment shows the result of this method. It removes the aliasing artifacts and much of the fringing.

The 3rd attachment shows it downsampled to 720p. Note that the artifacts are mostly gone and any postprocessing to remove the artifacts can probably be skipped. I also feel that this is the true reason why Pentax is pushing the 720p mode so much. It knows that the 1536x1024 shouldn't be used w/o further post-processing.

All in all, K-7 and K20 do exactly the same when extracting a video frame from the sensor. The K-7 just does it faster due to twice the channels.

Attachments:
  1. 400% crops from native 1536x1024 footage
  2. upper: 100% crop from native 1536x1024 footage,
    lower: reprocessed according to a method which a had published for the treatment of K20D burst video files.
  3. The same as above, but downsampled to 720p.

Last edited by falconeye; 06-15-2011 at 05:29 AM.
06-03-2009, 07:23 AM   #558
Senior Member




Join Date: May 2009
Location: Portugal
Posts: 112
Falconeye, you should work for pentax...

This 1536 x 1024 video mode is complitely worthless...
It shouldn't be implemented, it's not a standard and have less quality than
standard 720p.
Pentax:
make it 1080p supersampling for the next time, if you are going to stay seriously in video mode. (just look at the market emerging and draw your conclusions)
06-03-2009, 09:34 AM   #559
Forum Member




Join Date: Jan 2008
Posts: 55
I fully agree with NunoBarbosa, this mode is unuseable. It clearly show the artefacts (pretty disturbing indeed) and resolution is crappy as hell. It looks like when you resize an image with the worst algorythm, with no anti-aliasing applied, this is really awfull especially after viewing the canon video... they rushed the video mode just to mention it on specs sheet. pathetic. Have to save for 5d II now

06-03-2009, 10:54 AM   #560
Inactive Account




Join Date: Jun 2009
Posts: 3
My first post here.

I felt like in heaven for 48 hours at least. After seeing the first samples on Vimeo I really wanted the K-7. This last clip made me crash back to earth. The camera may be fantastic for close ups of foliage, oranges, dogs, faces, anything with soft, furry edges but the world also has edges ;-)

I'm waiting for a DSLR that scales correctly from the entire sensor area. There's a very small chance that this clip is not representive for the final product but I doubt it. If it is I'm sorry Pentax then you just had my attention for 48 hours.
06-03-2009, 02:55 PM   #561
Veteran Member




Join Date: Aug 2007
Location: Brazil
Posts: 376
Worse than the artifacts is the stuttering, is that frames repeating or what? Hopefully 720p looks better.
06-03-2009, 02:58 PM   #562
Veteran Member
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,867
Original Poster
Suggestion for K-7mk2 video specification ...

Because I do right now have a pretty clear idea of the workings of the K-7 video frame extraction, here are my suggestions to Pentax for their next update to the sensor:

1. Implement hardware binning, using a 3x3 array of sensor pixels to produce three color values or a 4x4 EXR array to produce 4 new bayer values. Below are two possible layouts for hardware binning, the second making use of the Fuji EXR idea:
[cf. attachment]
I wouldn't use the left variant (it creates fringing again). I rather would bin 4 or 5 greens and 1, 2 or 4 red/blues into a RGB triplet. The binning would always be symmetric around the center of a pixel and produce very clean colors.

2. The current 14.6 MPixels are more than enough to support "better than" 720p. Increase the #pixels to 5760 x 3840 (22.1 MPixels) to support a native 1080p resolution, noise-free, antialiased and crystal-clear. Or choose any compromise in between as a certain degree of upsampling is ok.

3. At 30fps, this architecture would require about 8 channels (rather than 4) of roughly the same speed each. The read-out would be after binning.

4. Offer MJPEG and 40MBit/s AVC codecs and 24, 25, 29.97, 30fps.


Such a video subsystem wouldn't require a lot of processing power but still equal the image quality which does true supersampling using brute force processing, i.e., rival IQ from the best of the best.

Last edited by falconeye; 06-15-2011 at 05:29 AM.
06-03-2009, 03:11 PM   #563
Veteran Member
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,867
Original Poster
QuoteOriginally posted by ricardobeat Quote
Worse than the artifacts is the stuttering
The stuttering is in the player only.

Even the 1536x1024 sample plays perfectly smooth on a PlayStation3. After resampling to 16 MBit/s AVC.

06-03-2009, 03:37 PM   #564
Veteran Member




Join Date: Dec 2008
Location: NYC
Posts: 943
QuoteOriginally posted by NunoBarbosa Quote


log in vimeo to download raw sample
yeah .. this not great at all

is it the K-7 for sure?... if it is and its not the final software they should take it of the web because this could damage sales

I already recommended the K-7 to a few colleagues and I hope they will not see this
06-03-2009, 03:52 PM   #565
Veteran Member
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,867
Original Poster
QuoteOriginally posted by redpigeons Quote
is it the K-7 for sure?... if it is and its not the final software they should take it of the web because this could damage sales
Well, it looks exactly as my analysis would have led to expect. Despite the critics I earned from some So, I guess it is genuine K-7 footage. The container details look original, too.

Because Pentax didn't solve this issue since the days of the K20D, I am pretty sure that it is too late to do anything about this now. It may actually be bound to the sensor hardware.

If Pentax decides to do anything about it, then it would be to remove the option from the firmware altogether. If Pentax thinks about it, then please, provide an option in customs menu to re-enable it, just like ISO 6400. With a bit of post-processing, the 1536x1024 mode is useful, actually.
06-03-2009, 03:57 PM   #566
Senior Member




Join Date: May 2009
Location: Portugal
Posts: 112
Calm down boys!

1536x1024 mode really sucks, but 720p is just fine for what it is, much better than 720p from nikon or canon 500d, or really bad compression for GH1.
If pentax provide us full manual controls (5d2) it will be a really good tool!
I will shoot a music video with one of this when it arrives, then i'll take my final conclusions.
06-03-2009, 03:57 PM   #567
Veteran Member
Torphoto's Avatar

Join Date: Mar 2008
Location: Trinidad W.I.
Photos: Gallery
Posts: 619
that sample is actually scarey bad but you never know until the final version is released.
06-03-2009, 04:05 PM   #568
Veteran Member
falconeye's Avatar

Join Date: Jan 2008
Location: Munich, Alps, Germany
Photos: Gallery
Posts: 6,867
Original Poster
Poll?

Shall we propose to Pentax in firmware 1.0 to hide enabling of the 1536x1024 mode in a customs setting just like ISO 6400?

In order to avoid careless critics from non-knowledgeable test magazines?

(always assuming that the quality itself won't change anymore)
06-03-2009, 04:11 PM   #569
Veteran Member
Torphoto's Avatar

Join Date: Mar 2008
Location: Trinidad W.I.
Photos: Gallery
Posts: 619
QuoteOriginally posted by falconeye Quote
Shall we propose to Pentax in firmware 1.0 to hide enabling of the 1536x1024 mode in a customs setting just like ISO 6400?

In order to avoid careless critics from non-knowledgeable test magazines?

(always assuming that the quality itself won't change anymore)
I think it's not a bad idea to do this, hide it in a custom function like the iso 6400 option unless it improves some.
06-03-2009, 04:16 PM   #570
raz
Senior Member




Join Date: Apr 2007
Location: Timisoara, Romania
Posts: 248
falconeye, my suggestion to pentax, for their future version is to have a fast enough sensor and a fast enough processor in order to take full size 4600x3100 pixel images at 30frames per second, then to scale it internally to 1080p or 720p. This would produce the best quality - in sharpens, noise, lack of antialiasing artefacts.

About the 1536x1024, if the final version of the camera would output video as the sample above, yes, they should better hide it in the menu and promote mainly the 720p which looks pretty good.
Reply

Bookmarks
  • Submit Thread to Facebook Facebook
  • Submit Thread to Twitter Twitter
  • Submit Thread to Digg Digg
Tags - Make this thread easier to find by adding keywords to it!
artifacts, hdmi, hdslr, k-7, lv, mode, movie, resolution, video
Thread Tools Search this Thread
Search this Thread:

Advanced Search


Similar Threads
Thread Thread Starter Forum Replies Last Post
K-5 video quality dileepkrp Video and Pentax HDSLRs 16 10-10-2010 12:45 PM
Is there a difference in video quality? USB transfer vs SD? Nflguy33 Video and Pentax HDSLRs 2 05-19-2010 09:53 PM
Is the K-x the same video quality as K-7? justtakingpics Video and Pentax HDSLRs 2 04-23-2010 10:43 PM
quality of lens for video karl79 Video and Pentax HDSLRs 17 01-28-2010 01:37 AM
Help...Very Poor Video Quality JesseY Video and Pentax HDSLRs 5 12-20-2009 12:08 PM



All times are GMT -7. The time now is 09:26 PM. | See also: NikonForums.com, CanonForums.com part of our network of photo forums!
  • Red (Default)
  • Green
  • Gray
  • Dark
  • Dark Yellow
  • Dark Blue
  • Old Red
  • Old Green
  • Old Gray
  • Dial-Up Style
Hello! It's great to see you back on the forum! Have you considered joining the community?
register
Creating a FREE ACCOUNT takes under a minute, removes ads, and lets you post! [Dismiss]
Top