I assume that DxO wants to keep the installed base and to grow that base. PL4 Elite DeepPRIME is superior as a “simple” one step operation to anything from the previous competitors, including rental workflow such as the Adobe products. However, if the above review is factual (I no longer rent Adobe and thus cannot verify the results shown in the above URL), DxO PL5 Elite or earlier needs the above to remain competitive. I do not need this for overall pixel resolution increase for most general images, but I do need this for detail in subjects that are still small in the original image (raw, D850, even with an 800 mm or longer 35 mm format lens).
DXO should be in a good position for this. Adobe’s approach is quite clever. By limiting the upscaling to a simple 2 times it makes the coding/UI changes as simple as possible whilst not competing with programs like Topaz Gigapixel for more ambitious upscales which are more suitable to rescuing tiny old jpg files etc.
In DXO the upscale process would seem to fit fit in the “output to file” tab. There are suggestions from users of other upscale programs that upscaling is best performed prior to image editing. Thus DXO’s upscale process would appear to nicely fit in with their current “Export as DNG with only Optical and noise corrections” option. Upscaling is also more successful with images that contain low noise which again is a good match for DeepPrimes capabilities.
A 2 times upscale ability would enhance DXO’s advantage with smaller sensor/lower pixel count cameras like MFT. If you have a 60Mp sensor the need for upscaling is marginal but if you have a 20Mp MFT sensor you get a 80Mp DNG which is worthwhile. MFT cameras like the Olympus EM1-iii can already produce in camera 80Mp files but the limitation is that any movement in the shot creates creates non pixel peeping distortions.
DXO’s DeepPrime already conditions users to needing a GPU for rapid processing and only seeing a small preview of DeepPrimes impact; the same would be true of a 2 times upscale.
DXO already have some experience of AI processing and are leaders in applying AI during demosaicing so if they can apply upscaling as part of the demosaicing along with noise reduction they again put themselves in a unique position.
The photo processing market is very competitive and prioritising development time is critical for any team. Upscaling is going to become a big topic (rightly or wrongly) but the benefit to DXO-Photolab would be far greater than for other companies due to the way it compliments DXO’s existing capabilities.
I have looked at the Topaz products. I have downloaded and expect to cancel the Adobe monthly rental after I see just how well Super Resolution works, and will be considering it again if nothing non-rented emerges. (There are many issues with rental software licenses, including eviction and non-access upon discontinuance of the lease. The three main reasons I upgrade a software package are: software defect fixes that are not incorporated into earlier major releases; major new functionality (e.g., DeepPRIME); support for later raw formats and/or bodies and/or lenses that are not backported to earlier major releases. Thus, if the functionality meets my needs (DeepPRIME was a significant improvement, hence worth the upgrade permanent license fee), I have no additional bodies or lenses to support (including new raw formats), I do not need to upgrade. Two points from Topaz Gigapixel: What File Types are supported? .dng .png .tiff .jpeg Most raw file types except CR3 ; by comparison, DeepPRIME only works upon raw files;
In our (Topaz) AI products, the framework should be DeNoise AI → Sharpen AI → Gigapixel AI .
Thus, my next attempt shall be to get rid of Adobe rentals (again), and try Gigapixel on the cropped final images from DeepPRIME, exported as TIFF (that should have been “fidelity” than JPEG). I am assuming in all cases that the AI technology being used is some variant of a multilayer backpropagation neural network with extensive training, not a software model of a neurosynaptic computer (hardware). Again, I would have two uses for this: salvaging old images (either say from a Nikon D2/D300 or scanned images from emulsion, much of which are Velvia); and for those subjects, such as birds, for which I simply cannot get as many well resolved pixels as are needed even with longish glass (say, 800 mm in 35 mm format). As an aside, there is something that I do not understand about this industry. There has been one dominant vendor which now has a rental only approach. If, say, Topaz has “solved” one problem, and, say, Dx0 another, why cannot these vendors mutually share (presumably including revenues) under license these technologies and thus more fully compete with a dominant vendor? It is not as if these were different market segments (say, printers) – all of these are post workflow vendors.
it is a good idea.
Nice improvement for older 6-8 Mpixel cameras. Combined with deepPrime, this features would give a new youth to old pictures.
A smart way to merge old photos with new ones.
Tried the following.
Converted a Raw( Nef ) to DNG in Photolab 4.
Opened the DNG in ACR and ran the Enhance/ Superresolution feature. It will run on a DNG from Photolab but their was an awful underlying geometric pattern to the result. This did not happen to the Raw file when opened directly in ACR and then enhanced. Interestingly on the DNG from DXO the enhance Raw details option is not available from the ACR enhance feature. Will now try on a TIFF from DXO and see if results are acceptable.
I tried ACR (using the same acronym as in the preceding reply post) on a trial basis (I object to rentals to get to my own work, with eviction upon non-payment of the rental, and no cap on what the rental cost will be in the future). My observations were similar to the above. I have since loaded Topaz Labs Gigapixel AI on a 30 day trial (not the 5 day from Adobe) working with a high resolution but rather tight crop (subject was small in the frame) JPEG image produced by PL4 DeepPRIME from a D850 NEF. (Using DeepPRIME before Gigapixel is the operational equivalent of the recommended Topaz workflow.) For a fair comparison, I used 4x pixel increase. The file sizes are 1222274 bytes from DeepPRIME, and 15622731 after Gigapixel, or a 12.8 fold increase. However, using a software “loupe” on the Gigapixel result shows a marked improvement in resolution (albeit, interpolated pixels) without artifacts (neither pixel, luminance, nor chroma) – Gigapixel “delivers” what ACR “promised”. Albeit one test, one image – but the result is encouraging. At some point, perhaps DxO and Topaz could negotiate an arrangement despite being competitors (as both are competing with Adobe), or DxO may invest in the research and development to produce a DxO practical pixel resolution increase application for workflow. Hopefully, neither will be acquired by Adobe or some other market “predator” and thus decrease innovation and support. For now, despite the additional (one time) expense, I plan to license Gigapixel, although I shall make a few additional tests (the image I used for this test is a difficult one – DeepPRIME did very well by comparison with the original NEF).
Don’t know weather or not it will hold up to your expectations but maybe you give “A Sharper Scaling” a try? It’s free software that works quite well - at least to my findings. But, to be honest, I’ve never delt with upscaling images to much… → http://a-sharper-scaling.com/
The main drawbacks of this software may be: - It doesn’t take RAW-files - it only runs on Windows - on Windows it needs an installed/activated .NET Framework 3.5
The results I’ve got from it look quite good.
What further development is planned? ››› Quality improvement: maybe. Speed improvement: unlikely. Support for 16 bits per channel: unlikely. A 64 bit version: unlikely. A MacOS version: out of reach. This is not a professional software. It’s not based on the right technologies for performing fast, handling large data, supporting multiple platforms. I have to compromise, it’s just a spare time project. End excerpt.
I fully appreciate that some may be using the above software application with success. The fact that the presented images compare well with a previous Adobe product may reveal something about the genius behind the application above or about the less than stellar performance of some Adobe implementations.
I don’t know about everybody else, but I find it very useful to have more than one solution to processing. Thus I have and use DXO (for many years), Adobe, Topaz Sharpen and Gigapixel, and a few others, notably Autopano Pro, which has never been bettered in my view though now defunct.
I don’t feel DXO need to focus on a “Super resolution” type of module: the output from DXO PL4 Raw conversion with DeepPrime is already superior to other demosaicing software, and this is more than sufficient to earn its place. I have Gigapixel and find it superior to Adobe Camera Raw’s Super Resolution, especially if used in “compressed” mode which is somewhat counterintuitive but reduces pixellation of fine edges such as hairs or foliage. Super Resolution is OK on very sharp images but poor on less sharp images. On my system it doesn’t seem much better then the older “enhance details 2.0”. Often you have to experiment, which is why it pays to have several alternatives. If DXO were to develop something like this, I would want to see it performing better than others and I think that would be difficult - after all, “AI” is inventing detail, not creating it.
I fully agree with your comments. Several points: I (and a few others I know) have discontinued Adobe due to the rental silent partner approach – I do not rent software applications. I did a trial rental of Adobe Photoshop with Camera Raw SuperResolution – it did not uniformly perform well. All interpolation techniques, including convolutions, and including sharpening, are “inventing pixels”, and thus are not suitable for medical, legal, etc., purposes without detailed analysis of the created pixels. However, these are useful for images presented to clients for aesthetic purposes. Finally, many working photographers spend more time in workflow than the original image acquisition, and thus the least amount of time to get a “saleable” keeper image is an advantage. As it stands, I export from PL4E DeepPRIME to Topaz Gigapixel when I need to because of the size of the subject of interest after cropping (e.g., using a NEF from a Nikon D850 with a Sigma 150-600 Sport and Sigma TC-1401, a bird may still be a small portion of a frame – and lugging a 600 4 plus 2x TC to the “middle of nowhere”, as I do not have a porter, simply is not feasible for me – 1200 with the 600-2x vs 840 with the 600-1.4x). The export from PL4E to Gigaresolution works, but every added step is added time in workflow. I have several custom (my own) presets in PL4E, and having a proper (not Adobe current) resolution increase internal to PL4E as a preset could (should? would?) save my time. Or, can export, etc., operations be part of a PL4 preset (in which case, my apologies for my ignorance)?
The DNG from DXO is a linear DNG, effectively a Tif file in a DNG wrapper. This is the same for LR/PS when you produce a pano/HDR etc.