[BigMackCam]

After some encouraging results with Topaz Labs' DeNoise, Sharpen and (to a lesser extent) GigaPixel AI suite which I bought back in April, I recently purchased their JPEGtoRAW AI software, as I own a number of JPEG-only cameras and was interested in its claimed abilities to remove JPEG artefacts, expand and smooth out the tonal range and recover detail. Whilst it worked reasonably on some files, it actually added artefacts to others, and after a few days I requested a refund under Topaz' no quibble 30-day refund policy.

My refund was issued within hours (well done, Topaz Labs) - but I also got a nice reply from a support rep with some very interesting information:

I just wanted you to know that we've incorporated the majority of the features of JPEGtoRAW AI into Gigapixel AI. The same functionality of JPEGtoRAW AI has been incorporated into Gigapixel AI's compression mode. You can use a 1 to 1 ratio if you don't want to upscale, and this will still restore editing capabilities. To clarify, none of the capabilities would be lost by using Gigapixel AI instead of JPEGtoRAW AI. Restoring editing capabilities is referring to the function of converting a JPEG to a RAW image. If anything, the current Gigapixel AI will be better quality as it has our new AI Engine in place.

Gigapixel AI can convert jpg or png files to DNGs or TIFFs. Gigapixel AI's other job is to resize and enhance low resolution images. However, if the resolution or size is not an issue or should be kept the same, then you would want to use the "scale" model and use the "custom" option with an input of '1.'

This will not upscale the image, and it will stay the same size. When saving, you will have the option to create the file as a DNG or TIFF for uncompressed editing capabilities. If you do want to resize or enhance, then you can do that all-in-one step.

Having played around with some of the same problematic JPEGs in GigaPixel AI using the "Low Resolution" model (which works best with JPEG compression artefacts) and x1.00 scale, I'm achieving much better results than I did in JPEGtoRAW AI, and with all of the benefits I'd originally hoped for.

One photo I had problems with in JPEGtoRAW was a moon-shot taken with my Panasonic DMC-FZ62 bridge camera. The resulting raw had way more and more-pronounced artefacts than the original JPEG image... However, GigaPixel AI did a far better job. Here's a 1:1 comparison - on the left, the JPEG... on the right, the raw from GigaPixel, with a couple of minor additional adjustments in Lightroom after conversion:



Not bad at all, IMHO This is not simply noise-reduction... I can get rid of the JPEG artefacts using Lightroom's luminance noise reduction slider, but it kills the detail. GigaPixel AI not only retains the detail, but actually improves on it...

TL;DR #1 - If you're thinking of buying JPEGtoRAW AI, buy GigaPixel AI instead. For one price, you get better JPEG conversion results plus an often-quite-capable (though not miracle-working and sometimes frustrating) resize tool.

TL;DR #2 - If have a back-catalogue of JPEG images and own GigaPixel AI, you can use it to produce much better image quality from those JPEGs - including significantly reduced artefacts, "restored" detail, greater tonal adjustment latitude without banding etc.

---

Last edited by BigMackCam; 05-15-2022 at 01:03 PM.

[robgski]

Thanks for the Tip, I did convert some images using the JPEG to RAW software, I will try it out in GigaPixel as well.
For me, TopazAI has been a great software suite.

[Mooncatt]

I love their programs and already have Gigapixel, but I haven't tried outputting to RAW with that one. I've tried outputting as RAW (albeit from RAW originally) in DeNoise and the DNG's were not usable in the least. The file sizes were as large as a TIFF (which they told me is essentially how it saved the DNG's), but the images always rendered at a low resolution thumbnail size that was heavily posterized. I saw this in Raw Therapee, but don't have anything Adobe to check too.

I don't really have need for this feature, but I may give it a try via Gigapixel just for my own curiosity to see if it's any different.

[BigMackCam]

Thanks for the Tip, I did convert some images using the JPEG to RAW software, I will try it out in GigaPixel as well.
For me, TopazAI has been a great software suite.

I like the Topaz AI suite, especially DeNoise and Sharpen... though I don't find them very good at guessing the best model and settings automatically, and it's not always the model you'd expect to work best that actually does. Then, when that model is selected, there can be areas of an image that require tidying up in something like GIMP or Photoshop. When they work well, they work brilliantly - but they don't always work as well as hoped. Still, I'm glad to have them in my arsenal

I love their programs and already have Gigapixel, but I haven't tried outputting to RAW with that one. I've tried outputting as RAW (albeit from RAW originally) in DeNoise and the DNG's were not usable in the least. The file sizes were as large as a TIFF (which they told me is essentially how it saved the DNG's), but the images always rendered at a low resolution thumbnail size that was heavily posterized. I saw this in Raw Therapee, but don't have anything Adobe to check too.

I don't really have need for this feature, but I may give it a try via Gigapixel just for my own curiosity to see if it's any different.

I haven't run into the low-res thumbnail issue you mention. Are you running the latest versions of the tools?

One good reason to save as DNG rather than TIFF - if possible - is that some raw development tools (RawTherapee is one) only allow certain tools to be applied to raw files, and will disable those tools for any other type, TIFF included...

---

Last edited by BigMackCam; 05-15-2022 at 01:30 PM.

[Mooncatt]

I haven't run into the low-res thumbnail issue you mention. Are you running the latest versions of the tools?

One good reason to save as DNG rather than TIFF - if possible - is that some raw development tools (RawTherapee is one) only allow certain tools to be applied to raw files, and will disable those tools for any other type, TIFF included...

I was running the latest version of DeNoise at the time when I tried it. I just updated again today and started working on a session I took yesterday, and tested one. I imported the DNG and let it save as DNG as well, and that seemed to open in Raw Therapee ok. So I'm guessing they did fix the issue.

I had switched up my workflow, running everything through RT first, exporting as a TIFF, then running through whatever Topaz programs I need, then finally to GIMP. This time I'm trying Topaz first and leaving as DNG's to see how things work out. I'm still too early in the process to see how things work out overall, but it looks promising so far.

[Mooncatt]

Well it seems only some of their programs are updated with newer RAW processing. I moved on to Sharpen AI 4.1.0 (the most current right now) and it is still giving me highly posterized images in Raw Therapee. Hopefully it gets the new RAW treatment as well. Either way, thanks for the info so I can reconsider my workflow.

[biz-engineer]

though I don't find them very good at guessing the best model and settings automatically, and it's not always the model you'd expect to work best that actually does.

The reason for that is the variety of image content and irregularities of lenses. If you use a macro lens with flat field, there isn't much difference in frequency response between corners and center, so the center isn't over-sharpened. Now , if you use a zoom lens wide open with some areas not in focus and smeared corners, how can the tool know what should be sharp and what's not. For instance, if you have a portrait picture with the subject in focus and a lot of background completely out of focus, AI known it is intentionally out of focus so it excludes what's out of focus for the analysis and sharpening. But now consider a landscape picture with the lens focused at the hyperfocal distance, with only a little bit of de-focus in the foreground , and a tiny bit de-focus at the horizon, Sharpen AI "think" it is an "all-in focus" type of shot but some areas aren't sharp and it tries to get it all sharp, and in that case the center will get too much sharpening and the foreground and far horizon still won't get enough sharpening. I don't see a problem selecting the sharpening mode manually, because I don't consider advanced sharpening something for processing large quantities of images. Usually, advanced sharpening is necessary for prints, so it is unlikely applied to a large quantity of images (because prints = money, sharpening is applied as a last processing stage, after up-sampling to 300 or 360 ppi, before printing) and most cameras these days deliver good enough images for small prints or web posts that don't require advanced sharpening, normal sharpening does the job quick and fine for 800x800pix images. So based on printing 25 large prints a year, I have no problem taking a few minutes to tweak the sharpening modes before uploading files to the lab.

If you expect a tool that brings your pictures to a good level of sharpness, automatically and without artifacts, DXO is a good place to start because it read the lens model from the image exif , finds sharpening/CA correction coefficients from the database. Best tool I've seen the get sharp jpeg (not overly sharp) free of aberrations down to the pixel. But DXO isn't as good as Topaz "super-resolution".

Then, when that model is selected, there can be areas of an image that require tidying up in something like GIMP or Photoshop

If you use Topaz own mask feature (Sharpen AI), you normally don't need to do any further processing.

[BigMackCam]

If you expect a tool that brings your pictures to a good level of sharpness, automatically and without artifacts, DXO is a good place to start because it read the lens model from the image exif , finds sharpening/CA correction coefficients from the database. Best tool I've seen the get sharp jpeg (not overly sharp) free of aberrations down to the pixel. But DXO isn't as good as Topaz "super-resolution".

It's not that I expect any tool to do it automatically... I fully expect to have some input in the matter; but if that tool is going to make suggestions, unless they're frequently correct or at least in the ball-park, I don't see the value in them. For me, Topaz AI suite's automatic suggestions are usually plain wrong - there's almost always a better model, and even for that better model the automatic levels are usually too punchy. This doesn't take away from the usefulness of the software, as I make my own manual adjustments - it's just a little irritating I'd rather it just asked me to pick a model rather than make the wrong guess 90% of the time...

If you use Topaz own mask feature (Sharpen AI), you normally don't need to do any further processing.

One of the "problems" I'm finding with Sharpen AI is where an image has detail that changes gradually in sharpness due to depth of field - for example, tall grasses in a field that are visible from near to far. The natural effect of that depth-of-field can be lost with Sharpen AI, and this isn't something that's easy to deal with through the mask feature. Of course, if you're shooting at apertures intended to provide maximum depth-of-field, this isn't a problem - it's an advantage; but when the aperture has been chosen to produce a gradual transition in focus, Sharpen AI seems - to me, at least - to sharpen out-of-focus elements too much, and the subtleties of the transition can be lost...

That said, as I already stated, I'm glad to have these tools. They definitely have their uses, and I'm sure I'll find increasing benefit in them... just not with every photo

[Dartmoor Dave]

Yes, Gigapixel is the way to go. I've been using it to convert the jpegs from my little Optio W30 to DNG with 2x upsampling and it's done a very good job of it. Even better, once you've converted the jpeg to DNG you can then apply a custom DCP profile to it in camera raw. So you get almost all the benefits of raw from a jpeg only camera. Of course nothing can really restore the data that was thrown away when the camera made the original jpeg, but this method does a great job of reconstructing what it might have been.

[biz-engineer]

For me, Topaz AI suite's automatic suggestions are usually plain wrong - there's almost always a better model, and even for that better model the automatic levels are usually too punchy.

I agree. On my version of sharpen AI, I also discovered that there is a bug in the sharpness & noise reduction sliders. If the auto-setting is 30 for sharpening and 15 for noise reduction, but the amount of sharpness and noise reduction aren't the same when I set the slider to 30 and 15 manually. It feels like the inner sharpening algorithm are good, but the top level software implementation is low quality.

---------- Post added 17-05-22 at 10:32 ----------

One of the "problems" I'm finding with Sharpen AI is where an image has detail that changes gradually in sharpness due to depth of field - for example, tall grasses in a field that are visible from near to far. The natural effect of that depth-of-field can be lost with Sharpen AI, and this isn't something that's easy to deal with through the mask feature

The problem with Topaz masks is that they aren't are gradual as masks in PS or GIMP, and sometimes when I draw masks in Sharpen AI it's doing some heavy processing in the background and freezing, whereas the brush masking processing in GIMP is smooth and real time. There's a lot of room for improvements.

[Dartmoor Dave]

One of the "problems" I'm finding with Sharpen AI is where an image has detail that changes gradually in sharpness due to depth of field - for example, tall grasses in a field that are visible from near to far. The natural effect of that depth-of-field can be lost with Sharpen AI, and this isn't something that's easy to deal with through the mask feature.

This is where using a fully featured editing app with layer masking comes in useful. I sometimes end up with several layers with different types or amounts of sharpening applied, using layer masks to limit each sharpening (or sometimes blurring) effect to the parts of the photo where I want it. The masking features built into apps like Sharpen AI don't come close to what you can do with a proper layered editing app like Photoshop or GIMP. For example, I use a layer mask with between 1000 to 2500 pixels of feathering to separate a sharpened ground from an unsharpened sky.

(Looks like I was typing a similar reply to biz-engineer's at the same time.)