[MarkJerling]

Check out this DxO user's high ISO example! You can click on the image to see it larger.

DeepPRIME ? Well done! It's fantastic! - DxO PhotoLab - DxO Forums

That's very impressive, indeed.

[Kiwizinho]

Check out this DxO user's high ISO example! You can click on the image to see it larger.

DeepPRIME ? Well done! It's fantastic! - DxO PhotoLab - DxO Forums

You've convinced me to at least download the demo. I have a couple of images that convinced me to buy a new camera body to get better high ISO performance, but even now I'm not sure if I can really achieve what I want without stepping up to full frame.
DxO demo won't cost me anything, and if it works it's considerably cheaper than a K-1 II!
Lightroom definitely couldn't handle what I needed.

[Kiwizinho]

I've started having a play with DxO's DeepPrime noise reduction, and might yet pay up for it, but have found a few issues. The Lightroom integration works, but has issues. If it didn't I'd be posting an example here, but unfortunately after doing all the adjustments for an image that wasn't useable with Lightroom alone, I couldn't get DxO to save the thing back to Lightroom (or at all), although another image did work.
The other thing is processing is extremely slow. I'm using a less than 6 months old Windows laptop with an AMD Ryzen 7 CPU, so not too much of a slug, but it takes several minutes to process an image with DeepPrime.
The amount of noise reduction while retaining detail is impressive, and I guess you've got to be prepared for a tradeoff between accuracy and performance, but this is certainly not something you want to be running the majority of your images through.
On the plus side, an IS03200 night landscape image from my K-50 that I thought was too noisy to be useable was markedly improved.
I'll have another go processing it and if I can get it to work, will post it.

[zkarj]

Yes performance is not high, but I'm using a pretty basic Mac mini with integrated Intel GPU with only 1.5GB of memory and I don't find it excessive. If I'm processing a bunch of pictures, I will just set them all exporting and go do something else.

I've used the Lightroom integration a bit though I don't think I will stick with it because I don't like the double-ups in Lightroom. I did find that sometimes the "Send to Lightroom" button reverted to "Export to disk" but the dropdown always has the option for sending an image to Lightroom. I also noticed that Lightroom can take a while before it updates its interface to show the newly received images. Something on the order of 10-20 seconds.

I should add that while I don't intend to use the Lightroom integration, I will still use Lightroom. I import new photos there, add keywords, write those keywords to the (DNG) file and then open PhotoLab for processing. That way I get the power of Lightroom's keywording (no other software compares for me) and the power of PhotoLab's ability to resolve sharp and clear images (no other software compares for me).

---

Last edited by zkarj; 11-04-2020 at 03:26 PM.

[zkarj]

PhotoLab 4 allows you to select whether it uses GPU or CPU for "DeepPRIME acceleration." I'd left mine set to Auto but after Kiwizinho's comment about CPU it got me to thinking, so I just did an experiment. It seems the Auto setting was choosing my (2018 model) 3 GHz 6-Core Intel Core i5 to do the processing based on what I saw in the tests below versus what I normally experience.

I set it to explicitly use the CPU and exported the example image I posted above (though this time it was my fully processed version) into a 3200px wide 90-quality JPEG. This took 2:01 to complete.

Then I set it to use my wimpy Intel UHD Graphics 630 1536 MB. It took 1:34 to complete and the computer ran a few degrees cooler. (I started from a bottomed out CPU/GPU state and 55º for both tests.)

I wondered whether giving it multiple photos would make any difference, even though exporting one seemed to tax all my CPU cores. So I gave it two different photos to process. Based on the results of that — 4:02 and 2:58 respectively — it seems not to make any difference.

So my wimpy GPU may not be so wimpy after all. I mean I never notice any issues when I'm doing pretty much anything except trying to get a flight simulator to run. DxO do say that screen refresh may be "jumpy" during exports when using a partially supported GPU and mine is marked as such.

[richandfleur]

PhotoLab 4 allows you to select whether it uses GPU or CPU for "DeepPRIME acceleration."

Adobe have really ramped up the GPU support lately. Long overdue but makes a massive difference, like night and day on my machine now.
It's been in use lately on Premiere video export but is now rolled more into the core application use components like scrubbing through the timeline.
Likewise with Lightroom which got this over a year ago now.

Happy customer here, and have found the recent releases for the last year or more have been solid, reliable and crash/bug free.

[zkarj]

GPU acceleration is a good thing, and arguably WAY overdue considering game writers have known the benefits for decades.

What I think could be very interesting in the future is the more we see machine learning the more I suspect Apple Silicon will come to the fore, as it will (almost certainly) include their neural network technology and I've seen what that can do. Pixelmator introduced their ML Super Resolution feature a while back on their "Pro" Mac app, stating that "GPUs are now fast enough to perform the work in the amount of time the user is prepared to wait." On my Mac mini (specs above) it's certainly not fast, but I'm happy to wait it out because of the quality it achieves (doubling or tripling resolution). However, more recently they added the same function to their iPad app and my 2020 iPad Pro is WAY faster at it because they get to use the neural engine built into the iPad's chip hardware. We should get our first look at Apple Silicon next Wednesday, about which I am very excited.

As a side note to these topics, Pixelmator's ML Super Resolution upscaling and DxO's DeepPRIME demosaicking/denoising are fantastic uses of machine learning that bring substantial benefits to lots of people in lots of situations. This is in contrast to certain other vendors' approach where things like the ability to (often poorly) replace a sky is deemed a worthy feature. A skilled Photoshop user can replace a sky easily, but they can't upscale or denoise anywhere near what ML has done.

[MetteHHH]

Kia ora!

Sorry for crashing your party from the wrong end of the planet, but I was told this was the place to find Kiwi advice...

My brother and his wife are going to spend the next two yearsin NZ, and I made a thread (link below) to ask for some local advice on Christmas shopping. If anyone has photo calendar experiences to share, I would really love to hear about it, thanks.

Kiwi advice on where to get a printed photo calendar i NZ? - PentaxForums.com

[MarkJerling]

Kia ora!

Sorry for crashing your party from the wrong end of the planet, but I was told this was the place to find Kiwi advice...

My brother and his wife are going to spend the next two yearsin NZ, and I made a thread (link below) to ask for some local advice on Christmas shopping. If anyone has photo calendar experiences to share, I would really love to hear about it, thanks.

Kiwi advice on where to get a printed photo calendar i NZ? - PentaxForums.com

All good. I've answered you in that thread.

[Kiwizinho]

GPU acceleration is a good thing, and arguably WAY overdue considering game writers have known the benefits for decades.

What I think could be very interesting in the future is the more we see machine learning the more I suspect Apple Silicon will come to the fore, as it will (almost certainly) include their neural network technology and I've seen what that can do. Pixelmator introduced their ML Super Resolution feature a while back on their "Pro" Mac app, stating that "GPUs are now fast enough to perform the work in the amount of time the user is prepared to wait." On my Mac mini (specs above) it's certainly not fast, but I'm happy to wait it out because of the quality it achieves (doubling or tripling resolution). However, more recently they added the same function to their iPad app and my 2020 iPad Pro is WAY faster at it because they get to use the neural engine built into the iPad's chip hardware. We should get our first look at Apple Silicon next Wednesday, about which I am very excited.

As a side note to these topics, Pixelmator's ML Super Resolution upscaling and DxO's DeepPRIME demosaicking/denoising are fantastic uses of machine learning that bring substantial benefits to lots of people in lots of situations. This is in contrast to certain other vendors' approach where things like the ability to (often poorly) replace a sky is deemed a worthy feature. A skilled Photoshop user can replace a sky easily, but they can't upscale or denoise anywhere near what ML has done.

I think the whole ML/AI thing has heaps of potential, and it's interesting seeing products come out that apply it to images captured by reasonable sized sensors. There's been plenty of application to smartphone images, as they're inherently constrained by tiny sensors and limited space to put anything bigger, but smart software processing is equally applicable to images captured by larger sensors. Better noise reduction is an obvious use, but I also wonder if it could be applied to something like smart depth of field enhancement for macro photography without needing to stack images. I'd imagine the amount would be limited, but even a stop or two of extra DOF would be handy for macro.

[zkarj]

Well look what just popped up in my news feeds...

The 'Caspian Sea Monster,' a Lun-class Ekranoplan, rises from the grave | CNN Travel

[richandfleur]

Quite excited watching the parcel tracking of my new (second hand) 55-300 PLM WR lens heading to NZ.

That should do me for quite a while, with the 16-85 WR plus this giving modern glass on the K-3

Will look to move on my older 55-30 DA lens to make way and offset some of the cost.

Have been keenly following the new K-3 mkiii? camera in development, but that's taken a very very long time (K-3 was released in 2013 and the mk ii in 2015).
In all honesty though, I'm not sure how this will stack up against modern mirrorless developments and capabilities, but we'll see.

I've been spoilt by face AF detection and powerful object tracking capabilities of even basic mirrorless cameras, which I can't really see DSLRs ever offering outside of live view/mirrorless mode.
Think it rather depends on your subject matter, and the Pentax is still my go to for serious landscape type work.

[richandfleur]

Some more from Rotorua recently:

[MarkJerling]

Lovely images Richard.

[Kiwizinho]

Some more from Rotorua recently:

Wow! Some actual photos! Anyone would think you own a camera!

Must find some to post, as I have actually done some photography, although this year has been a bit quiet in terms of what I've done photographically.

---------- Post added 11-25-20 at 10:01 PM ----------

Quite excited watching the parcel tracking of my new (second hand) 55-300 PLM WR lens heading to NZ.

That should do me for quite a while, with the 16-85 WR plus this giving modern glass on the K-3

You have the 16-85 WR? It's probably the one lens still significantly on my wish list as it seems to get good reviews and is WR which would be handy for me.

For similar zoom range without WR, I have a Sigma 17-70, and with WR I have a D-FA 28-105 which I guess is the full frame equivalent field of view, but I picked one up considerably cheaper than a new 16-85 so figured even though it doesn't go as wide by quite a bit, at least it's a reasonably useful zoom range with WR.