Luminous Landscape Forum
Raw & Post Processing, Printing => Digital Image Processing => Topic started by: plugsnpixels on August 16, 2019, 12:27:53 am
-
Yesterday Topaz soft-released version 4.3.1 of Gigapixel AI (https://topazlabs.com/gigapixel-ai/ref/5/), with the following features:
•Toggle in right panel to enable and disable Face Refinement
•Face Refinement will detect small faces (16x16px - 64x64px) and apply targeted improved upsampling
They are also aiming for 3X faster processing in CPU mode.
Attached is a screenshot of the new GUI with a 15-year old, 4MP image being treated with 4X magnification.
-
Another sample of some unsuspecting woman at Disneyland, also from about 15 years ago. I love the 6X results, especially on the nose. Too bad we can't use this treatment on ourselves!
-
However they still have not removed the ridiculous pixel limitation of 22,000 pixels in any dimension. I was promised by them that they would do so four updates ago.
I'm not using it until they fix it.
Victor
-
However they still have not removed the ridiculous pixel limitation of 22,000 pixels in any dimension. I was promised by them that they would do so four updates ago.
I'm not using it until they fix it.
Hi Victor,
I have not seen them making such a promise. They did say that they were going to look into that and see if they could relax those limitations somehow, by circumventing some of the inherent memory issues that exist. The limitations are there to prevent memory overflow errors, or file format limitations (i.e. TIFFs > 2GB or 4GB, depending on the TIFF-library used).
Anyway, you can already increase the maximum output size to 32000 pixels, by selecting 8-bit/channel output files instead of 16-b/ch. That's not a huge limitation, given that printed output usually uses an 8-b/ch print pipeline, and that printer drivers also have maximum pixel dimension/length limitations.
I do agree that a 22000 pixels limit in any dimension is probably too restricted, even if only looking at file-format, because that produces at a maximum something like just shy of 1GB. So a 31000 pixel limitation would not even exceed the 2GB limit, and that's only in the worst case of a square image.
But additional RAM memory-related restrictions can be imposed by GPU memory limits or software libraries, which would require a potentially significant rewrite of the way the software is structured to allow circumventing those limitations. I suspect that is the reason they haven't 'fixed' it yet. Limited demand and thus lack of an incentive to splurge money on allocating a lot of resources.
Cheers,
Bart
-
Yesterday Topaz soft-released version 4.3.1 of Gigapixel AI (https://topazlabs.com/gigapixel-ai/ref/5/), with the following features:
•Toggle in right panel to enable and disable Face Refinement
•Face Refinement will detect small faces (16x16px - 64x64px) and apply targeted improved upsampling
They are also aiming for 3X faster processing.
Attached is a screenshot of the new GUI with a 15-year old, 4MP image being treated with 4X magnification.
I opened my current version and the upgrade notice didn't come up like it usually does.....maybe too soon.
-
My previous release version offered the update a couple of days ago (Mac).
-
I opened my current version and the upgrade notice didn't come up like it usually does.....maybe too soon.
Not showing up here, either (U.S. mid-Atlantic), at the time I am posting this. Either it's still propagating to the distribution servers or Topaz has pulled the release.
-
OK, I just checked their website and it says "Current Version: 4.2.2". I am just a Future Man!
-
OK, I just checked their website and it says "Current Version: 4.2.2". I am just a Future Man!
Were you updating from a Beta release to a later Beta?
-
From the previous release version, via auto-update. I am awaiting clarification from Topaz at the moment and will let you know what I hear.
-
OK, it seems some Mac users are having some issues which should be resolved in a couple of days. On my MacBook Air at home I'm good with 4.3.1
-
OK, it seems some Mac users are having some issues which should be resolved in a couple of days. On my MacBook Air at home I'm good with 4.3.1
So they pulled it. I'm looking forward to the update — not just the performance improvement (always welcome, obviously) but also the face enhancement, which sounds interesting.
Thanks for checking and reporting back.
-
I know they're working hard on this release and it involves some additional tech not in the prior version, so we'll wait a bit for it to be sorted out.
In the meantime, here's another one of my before-and-afters using a 15-year old 4MP image. Nobody is safe... ;-)
-
While we wait: Here's Scott Kelby and Dave Cross from Photoshop World Vegas 2006, 4MP original.
-
I just checked and I must have gotten the upgrade because I"m on 4.3.1. I just gave it a quick try and if it's any faster, you couldn't prove it by me. Don't have much time right this minute though so I have to wait until later to give it a real test.
-
Thanks Dave, I believe the speed increase is aimed at CPU mode, which is the fallback depending on your GPU setup.
-
Thanks Dave, I believe the speed increase is aimed at CPU mode, which is the fallback depending on your GPU setup.
I must have missed that. I normally use GPU mode but I’ll try CPU and see if there’s a difference. Thanks.
-
The official release of Gigapixel AI version 4.3.1 was just announced. See the For Sale section (https://forum.luminous-landscape.com/index.php?topic=131819.0) for a double-discount if you don't already own it.
-
lightroom plugin?
-
lightroom plugin?
Hi Dan,
During the various Beta testing cycles several things have been proposed. The plugin functionality is one of the features that was proposed, if I remember correctly. But for the time being, Gigapixel AI is a stand-alone application.
Cheers,
Bart
-
I tried with raw files, and it often works very well. However I'm surprise they don't offer a dng output instead of a flat tiff, I would appreciate à sort of ai demosaicing, with upsizing.
PS : the 22000px tiff 16b output limitations becomes to be annoying for very large format.
Anyway topaz is making a good job, I also tested jpg-to-raw, even if the name is maybe not well chosen, the soft is making à very good job for removing compression artifacts and color quantization from a random picture.
If Adobe published à lot of articles about nn, they seem to be slow and cautious about incorporate them in Adobe cc. They seem to be more focus on the general public than on pro and photographers.
I'm still waiting for ai demosaicing, which will become the new Nice step in raw développement workflow for photographers !
-
Well, ‘faster’ may be nice but I would like to see that Topaz is applying ‘quality’ first. Still too much artifacts to be usable for professional and/or quality work. This is also applicable to the rest of the ‘AI’ pack.
Regards,
Jaap.
-
I fed the app a full-sized (6000x4000px) Fujifilm X-H1 raw file, converted to 16bit TIFF in Raw Photo Processor. Before the update to v 4.3.1, the enlargement to 2x took 17m13s. After the update, the same procedure took 17m20s. I'm going to guess the Mac version got no update in speed, even with the 1.3 GB update. There is very little observable difference, if any, between the two images. What is interesting is that if you use Apply Image ... Subtract with a scale of 2 and an offset of 128 between the two images you can visualize the areas of difference - more importantly, you can see a distinct grid pattern across the image, perhaps the tiles that the app uses to process the image?
Maybe one day the application will use both GPUs in my trashcan MacPro....
kirk
(the attached screenshot has had a very heavy curve applied to the difference image to make the results more easily visible).
-
The latest Gigipixel AI update (performed at 04-sep-2019) crashes during installation. Rebooting and/or re-installation did not resolve this. This occurs on a Win7-64 system. All updates in the past went flawlessly by the way.
Since there is no roll back functionality implemented, nor is it possible to download a previous working release from Topaz, I’m now stuck with a non-functioning Gigapixel.
I’ve initiated a support case, no feedback from Topaz yet.
So be warned…..
Regards,
Jaap.
-
The latest Gigipixel AI update (performed at 04-sep-2019) crashes during installation. Rebooting and/or re-installation did not resolve this. This occurs on a Win7-64 system. All updates in the past went flawlessly by the way.
Since there is no roll back functionality implemented, nor is it possible to download a previous working release from Topaz, I’m now stuck with a non-functioning Gigapixel.
I’ve initiated a support case, no feedback from Topaz yet.
So be warned…..
Regards,
Jaap.
It's likely a Windows 7 related problem and perhaps the new version has a conflict with some Win 7 DLLs that has not been addressed.
While this won't help your current problem, it does point out the necessity of doing an OS drive image on a weekly basis so that you can recover your system drive and associated programs prior to a failed installation. Most of the software solutions that do system imaging can be configured to do such backups and I usually retain six weeks worth just in case. Fortunately, I've not had to reimage yet (Windows 10 OS).
-
It's likely a Windows 7 related problem and perhaps the new version has a conflict with some Win 7 DLLs that has not been addressed.
Yes, could very well be. Make sure the latest DirectX drivers are installed, and the latest Graphics card drivers.
This latest version release focuses on speed-ups that depend on tweaking CPU/GPU processing and potentially 6th - 10th generation Intel CPU/GPU acceleration. Older or different (e.g. AMD) configurations might or might not be negatively affected. We're at the forefront of processing capability with Gigapixel AI, so Win7 is not ideal.
Also, does crashing still occur when restarting the application? I too experienced an application crash during beta-testing, but subsequent application starts after the initial crash were fine.
Cheers,
Bart
-
Hi Alan, Bart, thank you both for your reply.
In the end the root cause might or might not be related to Win7. That’s now under investigation at Topaz. However if Topaz wants to release a Win7 compatible product then this Win7 environment is with all its limitations the source to work from.
Yes, I do make a weekly image of the OS disk and its programs while the data is stored elsewhere. Only in critical situations I revert back to a weekly old system configuration, not for something like a failing Gigapixel install.
I mentioned the missing roll back functionality and I must say that Topaz supported me today by providing me a link to an installer of a previous working release. Kudos to Topaz!
It’s good to be aware that the installation procedure itself fails, not the execution of the program. So there is not yet a call from Gigapixel to the various drivers. My Win7 release (Intel i7 CPU) is up to date and so is my Nvidia driver for the GTX1070ti card. The crashing occurs each and every time I run the installer. Re-booting doesn’t change this situation. I never had any installation issues with Gigapixel, JPEGtoRAW, Denoise or Sharpen AI’s.
Let’s see and wait what Topaz support comes up with. Just for fun I’ll share the by the installer generated error with you, see the attachment.
Regards,
Jaap.
-
I still run Win7 Pro x64 on an aging laptop, and while I do not think the latest GigaPizel AI runs any faster for me, it sure is nice to know that it has been coded to use my CPU to its full potential.
-
I still run Win7 Pro x64 on an aging laptop, and while I do not think the latest GigaPizel AI runs any faster for me, it sure is nice to know that it has been coded to use my CPU to its full potential.
Yes, that's a familiar look, using all power it is allowed to get.
-
It seems like a lot of performance oriented apps are written to exploit the parallel processing on a GPU while leaving several, if not most of the main CPU cores idle.
It was reassuring to learn that this GigaPixel uses what is available.
-
I still run Win7 Pro x64 on an aging laptop, and while I do not think the latest GigaPizel AI runs any faster for me, it sure is nice to know that it has been coded to use my CPU to its full potential.
I don't see that on my workstation:
-
I don't see that on my workstation:
If you open the Resource Monitor, at the bottom of the TaskManager's Performance tab in Win10, you can see the utilization of individual CPU cores. I get 8 individual charts for the 4 cores + hyperthreading of my CPU processor.
In general, that's a useful thing to watch whether the CPU's are slowing down the task, and/or if the GPU is doing its thing.
The latest Gigapixel AI allows to tweak some parameters as well, but that is easiest tested by trying the different settings in preferences and timing the total operation for 1 test image.
-
If you open the Resource Monitor, at the bottom of the TaskManager's Performance tab in Win10, you can see the utilization of individual CPU cores. I get 8 individual charts for the 4 cores + hyperthreading of my CPU processor.
While true, I think that's beside the point of my post, which showed aggregate CPU usage over time for all 16 physical cores and 32 hyper threaded "cores" of my workstation.
My problem with looking at 32 individual charts all at once with no time component, just 32 bar charts dancing up and down is that I can't see the forest for the trees. When I use some programs like Helicon and use the parallel processing toolbox in Matlab with the right programs, I can see the aggregate CPU usage go to 100%.
I will look at the options in GigaPixel AI. Maybe it is refusing to use more than a set number of cores.
Jim
-
I will look at the options in GigaPixel AI. Maybe it is refusing to use more than a set number of cores.
It might also mean that your box is faster than the application can schedule subsequent sub-tasks, so some cores have to wait for other processes to finish.
Cheers,
Bart
-
It might also mean that your box is faster than the application can schedule subsequent sub-tasks, so some cores have to wait for other processes to finish.
When I tell it to be aggressive about GPU usage, it uses about all the cycles available, and it runs faster. It uses relatively few CPU cycles.
-
When I tell it to be aggressive about GPU usage, it uses about all the cycles available, and it runs faster. It uses relatively few CPU cycles.
Looking pretty decent! (as far as one could hope) !
-
I tested on a 16 bit TIFF file from Lightroom from a Sony A7r2 ( 7952 X 5304) with 2X enlargement.
With CPU with Intel Optimizations: 21 minutes 40 seconds
With GPU: 6 minutes 40 seconds
However, the colors of the poppies in the shot are noticeably different when stacked in Photoshop.
For example:
L * - 68 68
a* - 48 49
b* - 86 100
I haven't tried to determine yet which match the colors in the original shot. I see some difference in detail as well with the CPU processed shot being a bit sharper. I don't know whether my video card drivers not being up to date could be the problem. I am on Windows 10 with latest update, card is Nvidia Geforce GTX970.
-
Thanks for posting the interesting information.
-
I tested on a 16 bit TIFF file from Lightroom from a Sony A7r2 ( 7952 X 5304) with 2X enlargement.
With CPU with Intel Optimizations: 21 minutes 40 seconds
With GPU: 6 minutes 40 seconds
I gave it a try with an a7RII image and the same 2x setting.
With NVIDIA P5000 GPU: 4:50.
GPU disabled (still used 43% of GPU 3D cycles, though): 8:32.
CPUs: two Intel Xeon 6144.
Jim
-
I gave it a try with an a7RII image and the same 2x setting.
With NVIDIA P5000 GPU: 4:50.
GPU disabled (still used 43% of GPU 3D cycles, though): 8:32.
CPUs: two Intel Xeon 6144.
I should have mentioned CPU - Intel Core i7 6700 at 4.0 GHz. Did you see any color shift or difference in detail at 100% between the two methods? I noticed it immediately when I stacked them in Photoshop and turned the top layer off and on. That part is puzzling to me. I checked to make sure they both had the originally assigned profile, both did.
-
Did you see any color shift or difference in detail at 100% between the two methods?
I will check.
-
I should have mentioned CPU - Intel Core i7 6700 at 4.0 GHz. Did you see any color shift or difference in detail at 100% between the two methods? I noticed it immediately when I stacked them in Photoshop and turned the top layer off and on. That part is puzzling to me. I checked to make sure they both had the originally assigned profile, both did.
Hi John,
I also noticed a difference in detail. My Graphics driver doesn't qualify for the minimum hardware specs, so I run Gigapixel AI in CPU mode. During Beta testing for this version, I was able to run a test in GPU mode, but with frequent crashes (as expected). The tests that did finish, showed from better to much better detail in CPU mode.
This is something that I wanted to address with the folks at Topaz Labs after the beta cycle that focused on performance tweaking. I could not be sure that it was not limited to my specific setup. But now that you also notice it, there may be something that needs fixing in the GPU code.
I've just sent a support request explaining this. I'm not sure they can match my (incompatible hardware) but they probably can test with your hardware. It would be helpful if you could send them a support request as well.
Cheers,
Bart
-
I probably should update my graphics driver to see if that could be the problem. Later today I will post the Dropbox link to the Tif file that I used to see if other people see the same problem with that file. I will also put a tiff file with a small crop showing the differences in Photoshop
-
Here is the dropbox link to a small section of the TIFF file:
https://www.dropbox.com/s/cswn9z0829qzdpd/_DSC4007.tif?
Here is the dropbox link to the Gigapixel CPU vs GPU versions in separate layers with a few eyedroppers placed:
https://www.dropbox.com/s/ykdc1ah72vs6v9l/_DSC4007_output_CPU_vs_GPU.tif?
If anyone with different GPU wants to do the Gigapixel 2X enlargements to see if they get the same CPU and GPU results, feel free.
Edit: posted correct link showing CPU vs GPU.
-
The link to the file comparing CPU to GPU was wrong. It has now been corrected for anyone who wants to see the differences.
-
I updated the graphics driver to version 24.21.13.9924 dated 9/5/18 using "Update driver" from Device manager. No change so presumably this is not a driver issue.
-
Here's what I'm seeing. I don't think there are consequential differences. These are rezzed down from a 2x expansion.
-
At the suggestion of a user on the Topaz forum, I turned off Max AI and used GPU memory low or medium and got slightly BETTER results than CPU. Of note, using this method did not change the colors when compared to Preserve Details 2.0 in Photoshop (but essentially produced a cleaner version of Preserve Details 2.0). There was more subtle shading of colors in this GPU version than the CPU version, a subtlety that matched the Preserve Details 2.0. So, I get a better conversion which is faster. I think this will be my default setting although need to test on other images to make sure this isn't a one off result.
-
Anyone having problems with gigapixel and dealing with fog or images with Norton effect applied?
-
Anyone having problems with gigapixel and dealing with fog or images with Norton effect applied?
Hi Jim,
No obvious problems with fog. Anything in particular that we need to look for?
BTW, I'm running in CPU mode.
Cheers,
Bart
-
No obvious problems with fog. Anything in particular that we need to look for?
BTW, I'm running in CPU mode.
Bart,
I don't have gigapixel. Just wondered how the program handles foggy/hazy scenes. Wondering if it sharpens those areas, but I figure you could mask out the effect if it did.
-
Bart,
I don't have gigapixel. Just wondered how the program handles foggy/hazy scenes. Wondering if it sharpens those areas, but I figure you could mask out the effect if it did.
Scaled Fog remains as smooth as butter, unless there is some detail.
That detail is then rendered sharper but not necessarily more contrasty.
I'll try and find an example.
Cheers,
Bart
-
As 'promised' an example.
It's not a fog picture but a long exposure time one, some 30 seconds. That would cause a fog-like effect, which Gigapixel AI handles gracefully at a 6x upscaling. Of course the detail is a bit lacking at this scale, but it's higher than from a regular rescaling and perfectly suitable for high-resolution (600-720 PPI) printing.
Cheers,
Bart
-
Thanks for the example. Is anyone using the new texture feature in LR over Precision Contrast from Topaz now?
-
Thanks for the example. Is anyone using the new texture feature in LR over Precision Contrast from Topaz now?
From what I've seen of the texture feature in LR, it is something more like Topaz Precision Detail (which is like Texture on steroids).
Topaz Precision Contrast is Clarity, on steroids. ;)
Both functions are incorporated in the updated Topaz Adjust AI product, making that a nice one-stop exercise with some AI thrown in. But that's something for another thread.
Cheers,
Bart
-
FWIW I just upgraded GP and now have version 4.4.0.
-
FWIW I just upgraded GP and now have version 4.4.0.
Yes, that's the current version, but with some concerns about differences in sharpness between Acceleration modes.
Cheers,
Bart
-
I've only been using Gigapixel for a couple of months but after updating to 4.4.0 on my Mac I've run into a problem I haven't seen before. For some reason I'm getting very noticeable artifacts. I'm using Settings = Auto so is there something in Settings = Manual that might help this? It takes so long to process that I'd prefer to not just randomly try changing things if someone can point me in the right direction first. Thanks.
-
I've only been using Gigapixel for a couple of months but after updating to 4.4.0 on my Mac I've run into a problem I haven't seen before. For some reason I'm getting very noticeable artifacts. I'm using Settings = Auto so is there something in Settings = Manual that might help this? It takes so long to process that I'd prefer to not just randomly try changing things if someone can point me in the right direction first. Thanks.
Hi,
It looks like it's picking up noise as if it were details. Try increasing noise reduction, starting with the 'auto' settings and incrementing it from there.
You might want to combine a separate background and a foreground scaling, each with different noise and sharpening settings.
Cheers,
Bart
-
Hi,
It looks like it's picking up noise as if it were details. Try increasing noise reduction, starting with the 'auto' settings and incrementing it from there.
You might want to combine a separate background and a foreground scaling, each with different noise and sharpening settings.
Cheers,
Bart
Thanks for the tip, that's exactly what was wrong. I took that picture from a boat and was using a higher ISO than normal for me. When I went back to the original image and decreased the noise in it Gigapixel sized it up without the artifacts. Thanks again for your help.