Quick example of Pixel Shift Conversion between LR and Silkypix

[Paul2660]

The Pentax K1, has gotten some good and bad press lately, but one review on Dpreview over a month ago really started a negative chain reaction.  This review has since been corrected, but as with many things, first impressions tend to be the most important. 

I wanted to attach 3 images, from a test I was working on yesterday.  Good lighting, and a bit of wind.  The wind was not consistent and tended to hit different parts of my scene with more or less force.  What is so impressive to me is just how bad the LR conversion is vs the one from Silkypix.  If all you have is LR or ACR for the K1, you really need to look to Silkypix for now.  I have no faith that Adobe will attempt to revisit their current conversion for Pixel Shift on the K1, and I base this on the fact that Adobe never has fixed their less than stellar conversion for the Fuji X-trans files.  The problems with that conversion have been known about now for over 2 years. 

There are three images attached, the first one is LR's attempt on image 724.  Basically your first impression is "it's a totally wasted effort" nothing be done with this.  And if you stop at just the LR conversion you are correct.  The next image is the same raw pixel shift file, but with the Silkypix Developer Studio making the conversion and the difference is amazing.  It's like two different files but trust me, they are the same raw image.  This is not the Silkypix that comes with the K1 (that version is terrible, at least IMO) but the full version.  Note, currently the MAC version of Silkypix Developer Studio will error out on the Pixel shift files.  I have made them aware of this, but did not get a good feeling that they are terribly concerned about it, and will possibly address the issue at some later time. 

These conversions were done only for the effect to show the differences in blur recovery.  By default I feel Silkypix has too much sharpening applied and a bit too my saturation.  But a small price to pay for recovery of the image.  You can also see in first two files that the part of the tree just to the left of center, in the shade, is still having trouble, even with Silkypix, as it's blurred.  However I took several test files and in the very next frame you can see much more detail in this same part of the image. 

So for now, if you are using pixel shift or testing it, outdoors make sure you work with Silkypix Developer's Studio.  You can download the software for a 30 free trial and it's not inexpensive at $295.00.  It's a issue to take into consideration if you are looking at the K1 as the extra cost of this software takes the K1 into the $2,100.00 range.  Still a margin compared to other cameras in this class. 

I would like to see Pentax step and address this software issue, by at least giving a coupon or discount towards the full Silkypix software, as their version which ships with the camera might as well just be left out of the box. 

Paul C

[BernardLanguillier]

Interesting, thanks.

It is a bit sad to see that many of the interesting innovatins in digital photography these days (Fuji, Sigma Foveon, K1,...) are not supported by LR.

On a different note, I'd be interested some day in an explanation about how you select what camera to use among the many you own for a given assignement.

Cheers,
Bernard

[Paul2660]

Hello Bernard. 

I am still hoping that Phase One will release support for the pixel shift on the K1 sometime in the future.  C1 has one of the best X-trans conversions out there and adds all the other tools most are familiar with (similar to LR).  The LR conversions for X-trans vary, as they are excellent on people and buildings, but pale on landscape subjects, trees, rocks, (lichens) anything with a lot of finer details seem to be scrubbed.  In fact recently I am now realizing that a lot of this same issue creeps into the in camera jpgs from Fuji. 

I have also read the DxO plans to add K1 pixel shift support in September, and look forward to seeing how they do. 

Paul C

[torger]

The K1 DNG simply contain four images, one for each shift position. If there's been no movement you just need to layer them on top and you don't need to do any demosaicing at all. Super-easy. This is probably the only thing LR does.

The problem is if there's movement in the image, then they won't line up and you will get all sorts of artifacts. The solution is in theory easy, just pick one master image, say the first one, demosaic that and in all places there are movements blend in that master rendering.

The problem is then to make an algorithm to automatically detect where there are movements and where there is not.

If I had this camera I would ideally want in addition to a good automatic conversion also have an advanced tool where I could see how the automatic blending has been made and make it possible to make manual corrections to it.