Photoshop Enhance, cheaper than a high resolution camera.

[privatebydesign]

So I have been playing with the new feature in Photoshop 'Enhance', it is very impressive for uprezzing detail in images.

Here is a comparison of an original 20mp file that has been converted to an 80mp file, the first is the new 'Enhance', the second the best I could do in PS previously to get a 24" x 36" print. The third is the original 20mp image. All three are at 100%.

Pretty phenomenal in my book.






EDIT: I thought it might be useful to include the entire original image for scale.

 

[Joules]

Very interesting, thanks for sharing! Looks like a fun new feature to experiment with.

I'm curious if this works better if the lens is already 'outresolving' the sensor and to what degree the detail is true.

Edit:
Here is a nice read about the feature from one of the Adobe devs involved in its creation: https://blog.adobe.com/en/publish/2021/03/10/from-the-acr-team-super-resolution.html#gs.w4bdbi

It gets more interesting towards the end, where he posts side by side comparison of the DPR studio scene upscaled with a conventional algorithm to this new machine learning approach.

I find it noteworthy that this is supposed to be used with RAW files, so that the technique can perform the debayering and upscaling together for optimal results.

And that a capable GPU or M1 chip is recommended, so it looks like some laptops may have a hard time with this?

Unfortunately, he shared just one set of images from their training data and compared the result with the ground truth. Pretty close indeed, though seeing some examples from the test data set would have been nice. Just to get a better sense of how much real detail is in these results, and how much is an illusion of detail.

But there is a BIF example in his samples and it looks amazing. I'll give this a try shortly.

Edit 2: Do you guys this Canon knew about this and that's why they didn't bother with a greater than 20 MP sensor or TC compatibility recently?

 

[jd7]

So I have been playing with the new feature in Photoshop 'Enhance', it is very impressive for uprezzing detail in images.

Here is a comparison of an original 20mp file that has been converted to an 80mp file, the first is the new 'Enhance', the second the best I could do in PS previously to get a 24" x 36" print. The third is the original 20mp image. All three are at 100%.

Pretty phenomenal in my book.

Thanks for the heads up PBD! That sample result you've posted looks fantastic. I will definitely have to play with that feature at some stge.

 

[Joules]

So, after having played around with it a bit: Defintively a nice addition! It does not perform any miracles though. I tried it on some BIF images that were lacking detail and it didn't do anything to improve those. Garbage in, garbage out still applies here. Which in a sense is comforting to me, as it implies the neural network behind it isn't just filling in detail without any clues of what detail there should be.

Also tried it on a static bird with plenty of nice detail, but bad light. This creates a bit of a weird effect where the resolution and sharpness varies across different sections of the image. Some regions of the birds face remain virtually unchanged, while others are enhanced. A bit unnatural and overall no real benefit. Settings: 80D, Sigma 150-600mm 6.3 C, f/7.1, 1/500 s, ISO 500

But when it works, it clearly does so much better than using regular upscaling or even the 'manual focus shift technique' of taking a stack of handheld images and aligning them after upscaling. I was out a while ago and tried just that technique on some nice architecture, so I compared just upscaling, using the enhance super resolution and the stacking. The enhance comes away as the winner easily, resulting in a better image and taking much less processing time. Settings: 80D, Sigma 35mm 1.4 Art, f/4.0, 1/500 s, ISO 125

Or side by side:

For context, these are some massive crops! Here are the original images:


Edit: All the crops were first sharpened using the smart sharpen lens blur option for deconvolution, and have a tiny bit of unsharp mask sharpen apllied as well in an attempt to even out the playing field.

 

[eli452]

I hope you have a thick skin. Even with proof most, on this forum, will try to lynch you for this heresy

 

[Sporgon]

Certainly a welcome addition for those of us that want to print medium sized prints at 720 dpi.

 

[john1970]

Thank you for the great comparison. I am glad to see that software companies are utilizing AI to make processing better every day. The main reason I shoot RAW is that I know five years from now there will be improved software that will allow me to have better processing and output of the image than I do today.

 

[AlanF]

Topaz uses Machine Learning for Gigapixel upresolving software and for its sharpening. I compared PS Enhance with Topaz on a small crop of a Dunnock I took yesteday, which is a nice sharp image, albeit it small. Here is the original crop output from DxO PL4 with standard lens sharpening, then output without sharpening but upresolved 2x and sharpened with Topaz, and then PS Enhance. The PS is too slow on my MacBook pro with the 45 Mpx from my R5. I prefer the Topaz.

 

[AlanF]

And here is test from this morning's photos. I took a shot of a Little Owl from about 60m with the R5 and 100-500mm plus 1.4xTC. This is really pushing it because the image is so small - the owl is about 300x450px and the 1.4xTC softens the lens a bit. Topaz does a much better job here.

 

[YuengLinger]

Maybe the R6 will get a little more well deserved respect now, thank you.

 

[AlanF]

Maybe the R6 will get a little more well deserved respect now, thank you.

Nothing wrong with the R6 - it deserves respect without any need for software boosting. If you want more megapixels, the software that boosts the R6 by 2x also boosts the R5 by 2x.

 

[JPAZ]

I am intrigued by the changes in the industry. Camera technology is, of course, arguably improving. But, using AI, post-processing options are also growing. It seems to me that the "elephant in the room" (Adobe) is being pushed by the likes of Topaz and Luminar (and others) to up their game. We've got better options for selections, masking, up-rezing, noise reduction, and color manipulation than we've ever had. This is a good thing.

 

[privatebydesign]

I am intrigued by the changes in the industry. Camera technology is, of course, arguably improving. But, using AI, post-processing options are also growing. It seems to me that the "elephant in the room" (Adobe) is being pushed by the likes of Topaz and Luminar (and others) to up their game. We've got better options for selections, masking, up-rezing, noise reduction, and color manipulation than we've ever had. This is a good thing.

It’s a good point, Luminar does a much better job than Photoshop at sky replacements, the latest release of Luminar is simply staggeringly good at shy replacement even with reflections!

Topaz often does a better job of resizing images than PS too, but it does wrk me when people say Adobe don’t do anything for the subscription fees, and if I can do everything within one program, or suite, personally I’d much rather do that.

 

[privatebydesign]

Thank you for the great comparison. I am glad to see that software companies are utilizing AI to make processing better every day. The main reason I shoot RAW is that I know five years from now there will be improved software that will allow me to have better processing and output of the image than I do today.

Indeed, I’m going to look back at some of my favorite 1D images that I was always sad they were only 4mp, but even without resolution upgrades just the way newer software processes old files can bring new life into them.

 

[unfocused]

Thanks for calling attention to this. I've usually used One1 Resize, formerlly known as Genuine Fractals. It's very good, but this is appealing because it works inside of Camera Raw/Photoshop and does seem to yield good results.

 

[unfocused]

Adding...I tried it out on some photos I took yesterday at a soccer game. One of the frustrations of being a one-man band is that you can't be in more than one place at a time. I usually shoot from the opposing team's goal so I can get my players faces and the offensive action, but that means that anything I shoot on the other side of the field, such as our goal keeper blocking a shot, requires some severe cropping.

First, let me say that for some reason the 1DX III seems to tolerate these heavy crops much better than the 1Dx II. I have no idea why, but it may have to do with the new AA filter changes Canon introduced with the 1DxIII. That said, I can see using this feature in the future as my workflow requires that I produce one file for web use and another larger file for later print use in college publications.

I'm anxious to have my wife try it out on her songbird photos, which often must be radically cropped even with a 90D.

 

[R1-7D]

Topaz uses Machine Learning for Gigapixel upresolving software and for its sharpening. I compared PS Enhance with Topaz on a small crop of a Dunnock I took yesteday, which is a nice sharp image, albeit it small. Here is the original crop output from DxO PL4 with standard lens sharpening, then output without sharpening but upresolved 2x and sharpened with Topaz, and then PS Enhance. The PS is too slow on my MacBook pro with the 45 Mpx from my R5. I prefer the Topaz.View attachment 196285View attachment 196286View attachment 196287

Thanks for doing the comparison. I was going to ask how this PS Enhance feature compares to Gigapixel, which I use quite regularly with tremendous success.

To my eye, the Gigapixel version looks better; the PS version looks over sharpened, and the grain/noise isn't nearly as well suppressed.

 

[R1-7D]

Just for fun, here's a shot that I took when I was in the Cinque Terre back in 2018. It was an enormous pano stitch with my 5DsR - a 120 megapixel in total. I upscaled it using Topaz's Gigapixel to a 483 megapixel image.

The first image has a white arrow pointing to a man standing on a balcony way off in the distance. The second attached image is a blurry crop of the man from the 120 MP image. The third image is a crop of the man from the 483 MP image.

Pretty incredible, IMO.

 

[AlanF]

Thanks for doing the comparison. I was going to ask how this PS Enhance feature compares to Gigapixel, which I use quite regularly with tremendous success.

To my eye, the Gigapixel version looks better; the PS version looks over sharpened, and the grain/noise isn't nearly as well suppressed.

The noise is very well suppressed in the Topaz version because it was applied to the jpg produced by DxO PL4 with PRIME noise reduction. The PS Enhance just had noise reduction from the PS sliders. I am hopeless at noise reduction using PS and don’t know what settings to use to retain detail. I would have used Topaz Denoise but I don’t have it.

 

[Antono Refa]

Just for fun, here's a shot that I took when I was in the Cinque Terre back in 2018. It was an enormous pano stitch with my 5DsR - a 120 megapixel in total. I upscaled it using Topaz's Gigapixel to a 483 megapixel image.

The first image has a white arrow pointing to a man standing on a balcony way off in the distance. The second attached image is a blurry crop of the man from the 120 MP image. The third image is a crop of the man from the 483 MP image.

Pretty incredible, IMO.

Damn impressive.