[BruceBanner]

I had an interesting experience yesterday.
I tried for the first time to throw 15 DNG RAW files into PS for the stacking SuperRes treatment, things didn't go well at all. The calculation time was beyond ridiculous (in total over 1hr for the various stages to complete), compared to around 10mins I normally get with 4-5 files. Furthermore the waiting time really hammered the pc making it hard to even browse webpages whilst waiting (whereas with 4-5 files during the various waiting time stages (2-3mins at a time etc) the machine was more operable).
Then the kicker... I saw my C: just being utterly consumed in losing space. 15 RAW files at about 40mb each... PS temp files from working on these files.... 100gb! Yup, I needed to make sure I had 100gb free space on the disk drive to cope with this process... far out...

The icing on the cake was when everything was said and done the shot wasn't that great lol, so I abandoned any further editing hehe.

I think I'm coming to a couple of conclusions regarding SuperRes.

1) 4-5 DNG files is workable, many more than that and I (personally) am seeing too much of a tradeoff in terms of waiting/processing time vs convenience.

2) I feel there is a certain amount of 'luck' involved, meaning that if 1-2 of the shots in the 4-5 frames presented is really off and having trouble being aligned, then it ruins the rest of the shots.

3) And I lack the necessary PS skills to identify the frames responsible for being really outta whack and throwing the image off. Is it possible at some point in the process to take the image off to RawTherapee and get some Motion Correction/Pixelshift treatment be applied and RT removes the parts of the image that are contributing to bad alignment? Akin to how we normally treat Pixelshifted shots in RT?

4) The subject matter might matter more than Pixelshift shots. Shooting a scene with a lot of foliage, or a city scene with cars, basically anything where wind or movement might affect is going to look worse (perhaps) than taking the same shot with Pixelshift in a tripod and then visiting RT for motion correction (at least this much is true for me and my current PS skills in navigating the issue of motion with SuperRes). If you're snapping static things, or landscapes little to no motion in the frame, then I guess it's workable.

I haven't given up on the technique yet, but tbh I am having better consistent results with Pixelshifting and there's hardly a moment I leave the home without a tripod anyway so...

[BruceBanner]

Sometimes it works out really swell however



5 frame bracketed shot taken handheld (0.3ev value shift), combined in PS for intention of Super Resolution process. Came out rather well I think.

[stevebrot]

Sometimes it works out really swell however



5 frame bracketed shot taken handheld (0.3ev value shift), combined in PS for intention of Super Resolution process. Came out rather well I think.

Yes, very nice. Might you pull a full resolution snippet for our amazement?


Steve

[BruceBanner]

Yes, very nice. Might you pull a full resolution snippet for our amazement?


Steve

Sure, go here; https://farm1.staticflickr.com/970/42239570352_1b49319777_o.jpg

[jbinpg]

The increase in linear resolution method is likely derived from the process developed by the people responsible for processing the first Hubble images years ago. This process was dubbed the "drizzle" process. I use it whenever I can in my astrophotography but I am cognizant of two things;: individual frames need to be the best quality possible; and for each multiple of enlargement, you need a huge increase in frames for the process to work. Thus, computing time takes forever the more frames you throw at it. I never go beyond 2x drizzle. I would need to get a top of the line computer upgrade to even think about doing a 3x drizzle. So, in short, I don't see a big benefit of going beyond 20 frames with a 2x drizzle. The use of lots of DNGs and lots of compute time likely means there was some other processing going on before the drizzle could be attempted, likely debayering.

[BruceBanner]

The increase in linear resolution method is likely derived from the process developed by the people responsible for processing the first Hubble images years ago. This process was dubbed the "drizzle" process. I use it whenever I can in my astrophotography but I am cognizant of two things;: individual frames need to be the best quality possible; and for each multiple of enlargement, you need a huge increase in frames for the process to work. Thus, computing time takes forever the more frames you throw at it. I never go beyond 2x drizzle. I would need to get a top of the line computer upgrade to even think about doing a 3x drizzle. So, in short, I don't see a big benefit of going beyond 20 frames with a 2x drizzle. The use of lots of DNGs and lots of compute time likely means there was some other processing going on before the drizzle could be attempted, likely debayering.

I tried 15 DNG files, and PS wanted 100gb in processing temp file folder for those files, and things took forever, so yeah.

[normhead]

Meanwhile for an extra couple of minutes of processing with my K-1 pixel shifted images....
K-1 and DFA 100 macro WR and Pixels Shift.







.

[stevebrot]

Meanwhile for an extra couple of minutes of processing with my K-1 pixel shifted images....
K-1 and DFA 100 macro WR and Pixels Shift.







.

I played a little bit this week with super resolution this week on my K-3 and was left with a strong desire to own a K-3II.


Steve

[microlight]

Hehe, go on, you know you want to!

[BruceBanner]

Meanwhile for an extra couple of minutes of processing with my K-1 pixel shifted images....
K-1 and DFA 100 macro WR and Pixels Shift.







.

Nice shots Norm. With your Pixelshift shots are you doing much motion correction? Visiting RawTherapee afterwards etc or...

-----

Here's a shot from a morning or two ago, handheld 5 bracked shot, ev 0.3 shift, and then made 'Super Resolution';



I actually tried converting two of these shots differently, one shot I had the 'middle' exposure placed at the bottom of the layer of images before changing opacity and flattening levels, the other I left the middle exposure in the middle of the stack. I felt putting the middle exposure at the bottom as 100% Opacity worked best (both images were of course edited for better exposure, shadows and contrast etc, I'm just giving my opinion of which version came out better exposed with similar tweaking).

I have my Bracketing order currently set to Custom Menu 2, 11 Bracketing Order 2, with five images it gives this ordering; -2, -1, 0, +1, +2. So when I made the change for this pic here I dragged the 0 file out of the middle and placed it at the bottom, the order would have went in photoshop (from bottom to top);

20% Opacity +2
25% Opacity +1
33% Opacity -1
50% Opacity -2
100% Opacity 0

I'm now considering changing the order in the camera to another choice, so that I don't have to manually drag a frame down. I'm now dithering which mode to choose. I can have the following options (that keep the 0 exposure at the bottom of the stack for PS);

Option 1;

20% Opacity +2
25% Opacity -2
33% Opacity +1
50% Opacity -1
100% Opacity 0

Or Option 4;

20% Opacity -2
25% Opacity +2
33% Opacity -1
50% Opacity +1
100% Opacity 0

Hmm..

[BruceBanner]

Super(resolution) Sister;

If I have my bearings right this is me facing the first of the Three Sisters, being privy to only seeing one of the Sisters.
Shot was taken handheld in a bracketed HDR style compilied in a 'super resolution' way.

[normhead]

Nice shots Norm. With your Pixelshift shots are you doing much motion correction? Visiting RawTherapee afterwards etc or...

I don't use motion correction, if the subject has moved and the image i messed up, I just toss it. Or use the first image, and toss the other 4.

[LensBeginner]

Christ Pantocrator, mosaic in the apse of S. Miniato al Monte, Florence.
Stack from a 20-picture burst, smartphone.

[damianc]

I sometimes like to print quite large, and at times I worry (probably unnecessarily) about dropping dpi. Then it struck me that standard pixelshift files have the ideal data for superresolution images. It's possible to extract each RAW image (and upscale 200% using nearest neighbour) using Rawtherapee-cli. These can then be aligned using Hugin's align_image_stack then stacked using Imagemagick to produce a 16 bit TIFF ready to go back into Rawtherapee for the edit.
Basically a simple short script to do the job from a single pixelshift DNG.
Preliminary results look good though it is quite CPU and memory intensive, but it does turn the K1 into a 144Mp camera (for static subjects).

[maw]

It's possible to extract each RAW image (and upscale 200% using nearest neighbour) using Rawtherapee-cli. These can then be aligned using Hugin's align_image_stack then stacked using Imagemagick to produce a 16 bit TIFF ready to go back into Rawtherapee for the edit.

The solution that you propose is theoretically valid and I have already experimented with it, but the alignment is the result of four images, each filtered through an RGGB channel,
and this can only be achieved if you have a static image.

But with images of people or objects in motion (typical foliage), you can't stack them, except to lose those small details due to deviations in the movement itself.
In short, you don't lose the range of RGB colours required for a good chromatism.