[stevebrot]

1) By me not doing this stage and doing the Create Smart Object>Stack Media etc, does this take care of the '8. Flatten the layers' stage, or does this still need to be applied either way you tackle it. And how does one 'flatten' the layers?

No. Flattening the layers = merging the stack to a single background layer that may be easily saved to disk at well under 1GB. You want destructive editing and you want to vary contribution by opacity. Set up the layer stack as simple raster imports. This is not a job for smart objects. It truly is done as a "Layers 101" project.

Follow the instructions

P.S. If you want to integrate HDR (merged bracketed exposures) with Super Resolution, it may be best to get the basic SR flow working first. Contribution from clipped pixels are noise by definition.


Steve

---

Last edited by stevebrot; 05-08-2018 at 04:07 PM.

[BruceBanner]

No. Flattening the layers = merging the stack to a single background layer that may be easily saved to disk at well under 1GB. You want destructive editing and you want to vary contribution by opacity. Set up the layer stack as simple raster imports. This is not a job for smart objects. It truly is done as a "Layers 101" project.

Follow the instructions

P.S. If you want to integrate HDR (merged bracketed exposures) with Super Resolution, it may be best to get the basic SR flow working first. Contribution from clipped pixels are noise by definition.


Steve

Thanks Steve. I've tried the HDR version (the valley shot before, tho it was created using smart object>media stack), I think it's ok? I might try the multiple bracketed shots with 0.3ev at different exposures to cover more range (so instead of 5 layers, 10 or 15, ie 2 or 3 bracketed shots with still 0.3ev increments and each bracketed shot at slightly different exposure from the last).

[microlight]

Hi Bruce.

Responses in order:

- I don't use LR as I don't need the cataloguing feature, and Camera Raw has all the bells and whistles. My PP is done in ACR, and with 4 or 5 images for super resolution, all get edited simultaneously. So in other words, I do all my editing before the super resolution conversion, although I suppose you could also edit the output TIFF, although I haven't done it this way. I don't think that this is one of the critical steps in the process.

- Don't skip the up-sampling to 200% - this is critical to the process. Ian Norman in his Petapixel article explains that the averaging process "will allow us to resolve detail at up to 1/4 of our original pixel size; so when we upsample, we increase the image to 4 times its original size". What this means to me is that increasing the image size gives the headroom necessary to increase the resolution - otherwise there's no benefit to doing this.

- I don't see the advantage of the Smart Object route (even though it only takes literally a couple of seconds on my PC for the conversion) as it takes very little time to adjust the individual layer opacities. Yes, for five images, the percentages you quoted are correct. This is a more intuitive step for me, without going throught 'black box' Smart Object.

- Flattening is the third critical step, after the upsampling and averaging, as otherwise your output TIFF is just too big, as you've found. It's under the 'layers' menu, at the bottom of the list, and is necessary even if working with a Smart Object, as the single object still has a stack of layers in it, as Steve noted. He's also correct about destructive editing being necessary, but it doesn't matter here because you're effectively doing it on a new file tab in Ps, and can therefore even remove the four or five unchanged DNGs that you originally imported into Ps as they're not needed any more.

So, bottom line is that you have to minimally carry out the upsampling, averaging and flattening to get improved resolution at a manageable size. I tend not to re-sample to 100% but keep the 200% TIFF as that's essentially my 'RAW' file for a picture taken using this technique. While slightly larger than a pixel-shift file, it's not something I do with every photo; in fact I only use it as a pixel shift replacement when I haven't got my tripod with me, so the extra storage space isn't an issue on a 2TB drive dedicated to photos. Hope this helps.

I hadn't thought of using it combined with HDR, but first thoughts are that it may not work, or if it does then the correct exposure may need to be on the bottom of the stack as this is the one with 100% opacity. As Steve indicates, get the super resolution bugs sorted out first, and once you have a reproducible process, suck it and see with HDR, it can't do any harm.

[LensBeginner]

I can't really edit a 20-layer stack / 20-images smart object, so as soon as I'm done, I flatten the whole thing. I still have the originale though.

Import in series to smart object with auto-align / Average is the way to go for me, saves a whole lot of time, instead of having to open every single file, select all, auto-align, and set the opacity of 19 layers.

YMMV, but I couldn't do it any other way.

[BruceBanner]

Hi Bruce.

Responses in order:

- I don't use LR as I don't need the cataloguing feature, and Camera Raw has all the bells and whistles. My PP is done in ACR, and with 4 or 5 images for super resolution, all get edited simultaneously. So in other words, I do all my editing before the super resolution conversion, although I suppose you could also edit the output TIFF, although I haven't done it this way. I don't think that this is one of the critical steps in the process.

- Don't skip the up-sampling to 200% - this is critical to the process. Ian Norman in his Petapixel article explains that the averaging process "will allow us to resolve detail at up to 1/4 of our original pixel size; so when we upsample, we increase the image to 4 times its original size". What this means to me is that increasing the image size gives the headroom necessary to increase the resolution - otherwise there's no benefit to doing this.

- I don't see the advantage of the Smart Object route (even though it only takes literally a couple of seconds on my PC for the conversion) as it takes very little time to adjust the individual layer opacities. Yes, for five images, the percentages you quoted are correct. This is a more intuitive step for me, without going throught 'black box' Smart Object.

- Flattening is the third critical step, after the upsampling and averaging, as otherwise your output TIFF is just too big, as you've found. It's under the 'layers' menu, at the bottom of the list, and is necessary even if working with a Smart Object, as the single object still has a stack of layers in it, as Steve noted. He's also correct about destructive editing being necessary, but it doesn't matter here because you're effectively doing it on a new file tab in Ps, and can therefore even remove the four or five unchanged DNGs that you originally imported into Ps as they're not needed any more.

So, bottom line is that you have to minimally carry out the upsampling, averaging and flattening to get improved resolution at a manageable size. I tend not to re-sample to 100% but keep the 200% TIFF as that's essentially my 'RAW' file for a picture taken using this technique. While slightly larger than a pixel-shift file, it's not something I do with every photo; in fact I only use it as a pixel shift replacement when I haven't got my tripod with me, so the extra storage space isn't an issue on a 2TB drive dedicated to photos. Hope this helps.

I hadn't thought of using it combined with HDR, but first thoughts are that it may not work, or if it does then the correct exposure may need to be on the bottom of the stack as this is the one with 100% opacity. As Steve indicates, get the super resolution bugs sorted out first, and once you have a reproducible process, suck it and see with HDR, it can't do any harm.

Excellent reply, thanks that's really helpful.

When I was first alerted to Super Resolution it was from a fellow I follow on flickr (a fellow Pentaxian) but also he is in the same Pentax FB group as well. I felt I pestered him enough so started this thread here to get more responses (and also alert others to this technique/idea, just in case they (like me) weren't really aware of it). He did hint that he uses HDR Stacking a fair bit, essentially it is always wise to capture as much RAW data from the site being photographed as possible, and bracketing does just that. But yeah, I have not pestered him to ask exactly his work flow in this regard (I get the feeling he might be a Darktable/RawTherapee kinda guy and therefore he can't completely fully advise this idiot LR/PS user). Anyway here's his page; ShinyPhotoScotland | Flickr I don't think he hangs out here?, but his work is of a high standard if you ask me, I get the feeling he knows what he's talking about (super pentax nerd ) heh.

I have never had much success with handheld bracketing/hdr. Really I am hoping that slight bumps in exposure (0.3) will help salvage more from the scene, whilst also carrying off that non 'hdr' look (due to the opacity layer thing rather than how a typical hdr merge shot would present, which I am not overly fond of. I tend to use ETTR principles when doing landscape work and protect the cloud highlights etc...). I aim to test this out over the weekend, heading to some more lovely Autumn gardens, I shall take my 5 bracketed shots and convert them twice, once to a regular hdr shot (perhaps processed in either LR's stock hdr merge or that Nik Collection thing) and once with a Super Resolution process, see which one comes out looking the winner. Because I really like the idea of being able to walk around without the tripod and grab HDR 'pixel shifted' shots

I realise I am trying to run before I can walk here, but I just like to get the answers first before wasting time and shots on edits where a simply tweak here or there could actually produce a much better end version. So with HDR bracketed layers, the bottom needs to be the 'correct' exposure, do you think it matters much where the other layers may lie? Because it's a K-1 I'm thinking the brightest shot as the top layer (weakest opacity), I always find shadows easy to salvage on the K-1. If the final version of the HDR tiff file looks quite dark I'm thinking that's probably a better start to the editing process than the other way around?

Perhaps an order like this;

+ 0.7 (20%)
+ 0.3 (25%)
- 0.3 (33%)
- 0.7 (50%)
0 (100%)

How does that look in theory?

In future, whilst standing there and taken the bracketed shot, without reframing I may try hitting the +/- ev button, turning the dial a notch and firing off a second round of bracketed shots at a different exposure and also combine these to the stack to create a 10 layer (tho that will take practice I am sure to keep the hand pretty steady on the scene and not deviate much, far more advanced and will leave awhile before attempting those, but the thought had crossed my mind).

Thanks again for the informative and detailed reply. I hope to give you some 'physical' results over the weekend.

I can't really edit a 20-layer stack / 20-images smart object, so as soon as I'm done, I flatten the whole thing. I still have the originale though.

Import in series to smart object with auto-align / Average is the way to go for me, saves a whole lot of time, instead of having to open every single file, select all, auto-align, and set the opacity of 19 layers.

YMMV, but I couldn't do it any other way.

I'm not sure I understand much of what your wrote here

[LensBeginner]

I'm not sure I understand much of what your wrote here

I only have 8 GB RAM, so I have to flatten when I use a large amount of layers. I wrote in a post some pages ago that I routinely do with 20-frames bursts consisting of 16 MP images from my cellphone, especially indoors or at night.

The second point is, I use File > Scripts > Load Files into Stack, and check the two options (Attempt to automatically align source images, and Create Smart Object after Loading Layers), because this does a lot of steps in just one.
Then I go Layer > Smart Objects > Stack Mode > whatever Media Aritmetica translates to in English... should be Mean.

Then I flatten, and I'm done.

[BruceBanner]

I only have 8 GB RAM, so I have to flatten when I use a large amount of layers. I wrote in a post some pages ago that I routinely do with 20-frames bursts consisting of 16 MP images from my cellphone, especially indoors or at night.

The second point is, I use File > Scripts > Load Files into Stack, and check the two options (Attempt to automatically align source images, and Create Smart Object after Loading Layers), because this does a lot of steps in just one.
Then I go Layer > Smart Objects > Stack Mode > whatever Media Aritmetica translates to in English... should be Mean.

Then I flatten, and I'm done.

I think you want to be careful with that 'Attempt to automatically align source images' during the initial stack, the article I link mentions (I think) that it's a BAD IDEA to incorporate this step at this point, you want to Align later and in a specific way. But hey whatta I know

[LensBeginner]

I think you want to be careful with that 'Attempt to automatically align source images' during the initial stack, the article I link mentions (I think) that it's a BAD IDEA to incorporate this step at this point, you want to Align later and in a specific way. But hey whatta I know

Always worked for me. Read the PetaPixel article, but I also found little difference in the end result in doubling the linear resolution beforehand (tried both ways). So...

Anyway, you could still leave that unchecked, and use the import function only to batch import. It still saves time.

[stevebrot]

Because I really like the idea of being able to walk around without the tripod and grab HDR 'pixel shifted' shots

I have to admit this is very intriguing, but still believe that it is best to pin down the basics on one before trying to munge the two together. HDR is not a simple merge and usually involves selective masking on one or more layers.* That being said, I suspect that a five-image hand-held HDR bracket shot in "one push" (burst) probably provides some of the benefits of a super-resolution shot, though not equally for all areas of the frame.

I have been been looking for a good subject for a Super Resolution test with my K-3 and will report back if successful.


Steve

[BruceBanner]

I have to admit this is very intriguing, but still believe that it is best to pin down the basics on one before trying to munge the two together. HDR is not a simple merge and usually involves selective masking on one or more layers.* That being said, I suspect that a five-image hand-held HDR bracket shot in "one push" (burst) probably provides some of the benefits of a super-resolution shot, though not equally for all areas of the frame.

I have been been looking for a good subject for a Super Resolution test with my K-3 and will report back if successful.


Steve

But it might if you do the burst bracketed burst 4-5x and now have 20-25 layers to work with like (ie a typical Super Res is 4-5 same exposure frames to work with, you'd have that and more if doing the bracket burst multiple times). Would we not?

[stevebrot]

the article I link mentions (I think) that it's a BAD IDEA to incorporate this step at this point,

Read the PetaPixel article...

This ^ ^ ^

Too much information from too many sources ==> an amazing amount of noise in the channel.

Much as it might seem good to shortcut the learning phase, I don't think it is possible to sort through the chaff without gaining expertise by doing. When confident that the general recipe works (detailed in the Peta Pixel article), the images used for success may be reused as a basis for comparison with the results from other "recipes" or incremental changes to the base recipe.

One more thing...Don't worry about editing the 20 base images before the SR result is generated. Treat each exposure as 5% of the capture data needed to generate the Super Resolution result. Make believe this is all happening in-camera. Edit the composite, not the parts.


Steve

[stevebrot]

But it might if you do the burst bracketed burst 4-5x and now have 20-25 layers to work with

I suspect there "might" be a good reason why HDR is specifically denied for both regular PSR and dPSR images. Let this HDR idea set until you manage to complete a 20-exposure SR image with expected results at the expected file size.


Steve

[BruceBanner]

I suspect there "might" be a good reason why HDR is specifically denied for both regular PSR and dPSR images. Let this HDR idea set until you manage to complete a 20-exposure SR image with expected results at the expected file size.


Steve

Yes Dad... <stomps off kicking dirt>

[LensBeginner]

The proof being in the pudding, I'll post a couple of crops as soon as I can, so I can show you what I'm talking about

Aaah ok, I managed from my smartphone!
Detail (picture is much larger) from single shot, detail from fusion, 100% crop.

[microlight]

Input image quality is an important consideration here. I just tried the smart-object as well as the percentage-opacity method at both 100 and 200% for 2 x 5-shot cellphone sequences that I did a couple of days ago, as well as the 5-shot K-3II sequence that I posted on the previous page. The results confirm what everyone's found, in fact:

- The smart-object and percentage-opacity methods seem to work equally as well, except when there's movement between the frames where the smart object processing/median filter does better than the opacity method, although neither is perfect at dealing with the movement anomalies.
- For the cellphone sequences, upscaling to 200% produces little if any significant difference to 100% due to resolution limitations, certainly using 5 images.
- For the K-3II pictures, there is a definite difference between 200 and 100% using the sequence I put on the previous page, in that the non-upscaled image is both more pixellated and noisier than the upscaled one.

Processed with no upscale, 400% crop:


Upscaled to 200%, 200% crop:


You might get better results using 20 images from a cellphone, but it doesn't look like any more than 4 or 5 at DSLR resolution would be needed.

I mentioned movement between images; as with pixel-shift, there is blurring, but this can easily be edited in Photoshop (certainly using the opacity method) to remove any small areas of blur.