New iWE. Enhanced demosaicing; is it the game changer?

[Serguei Palto]

Newer version of my wave RAW editor is available at https://drive.google.com/drive/folders/1Hvx3-j_gHf5_2QMl78z0NO3JmZ_F4f3M?usp=sharing

In this new version of iWE the wave demosaicing algorithm is shifted to the GPU platform. Moreover, the GPU version is significantly enhanced compared to the older CPU version.

Now in the GPU version I have realized my old idea of getting the maximum from a color sensor in terms of the resolution, in order to make it closer to the black-white sensor of the same pixel count.

It is well known that, for example, the 20 Mpx sensor has only 10 M of the green pixels and 5M+5M blue-red pixels. Even enhanced demosaicing methods can't bring the resolution in the colored image that is compared with one from the 20Mpx black-white sensor. It means that 20 Mpx color sensor is not actually 20 Mpx sensor.

Nevertheless, spectral algorithms implemented in iWE software allow for significant step in direction of extracting fine-details information, which is still captured by a color sensor, but distributed over the pixels of different colors, so it is indeed hard to extract it by standard demosaicing methods.

Below are examples of using iWE algorithm in comparisson with one implemented in Adobe Raw Converter. In the both cases the same RAW file (p1000015.RW2; DC-G9, ISO200 from dpreview studio) has been processed with no sharpenss and denoising. The two output jpg-images were then resized by 2X and croped using the same linear interpolation in iWE, and their crops are shown below (#1 for Adobe, #2 for iWE).

The last two examples (#3, #4) are for the real life shot, that is processed using default settings in both Adobe PS and iWE with sharpness and denoise set to zero in the both cases. The iWE output after the wave demosaicing is naturally sharp due to 2X wider spectral band compared to an ordinary demosaiced image, but this "extra-sharpness" can be easely suppressed (if necessary) using a low-pass filter implemented into iWE at the demosaicing level.



[ATTACH alt="#1. Adobe RAW Converter demosaicing. This crop from 2X resized image I got using the Adobe PS 2021 for "p1000015.RW2" file from dpreview studio."]3037974[/ATTACH]
#1. Adobe RAW Converter demosaicing. This crop from 2X resized image I got using the Adobe PS 2021 for "p1000015.RW2" file from dpreview studio.




#1. The same RAW-file as in #1, but processed using the iWE wave demosaicing algorithm.




#3. Raw image processed in PS.




#4. Raw image processed in iWE using newer GPU-implemented wave demosaicing.

 

[kalisti]

Wow, looks like you're making really good progress on this.

I'm very impressed with the examples, yours look better than the PS versions imo, and I love my PS XD

 

[Serguei Palto]

Wow, looks like you're making really good progress on this.

I'm very impressed with the examples, yours look better than the PS versions imo, and I love my PS XD

Yes, imo as well . In case of iWE each of the color channels contains high-frequency information registered by all the pixels. For example, in each row of the sensor only half are green pixels, but in my case the green channel contains the high-frequency information registered by all the pixels including the non-green pixels. So, in image #2 you can see the text information which is really hard (if possible at all) be extracted by demosaicing methods based on space-domain interpolation algorithms. The disadvantage of my method - it is very costly in terms of computing resorces.

 

[kalisti]

Wow, looks like you're making really good progress on this.

I'm very impressed with the examples, yours look better than the PS versions imo, and I love my PS XD

Yes, imo as well . In case of iWE each of the color channels contains high-frequency information registered by all the pixels. For example, in each row of the sensor only half are green pixels, but in my case the green channel contains the high-frequency information registered by all the pixels including the non-green pixels. So, in image #2 you can see the text information which is really hard (if possible at all) be extracted by demosaicing methods based on space-domain interpolation algorithms. The disadvantage of my method - it is very costly in terms of computing resorces.

Eye opening for sure!

Does it utilize lots of cores or lots of frequency?

Thanks for the extra details, love the work so far

 

[Serguei Palto]

Wow, looks like you're making really good progress on this.

I'm very impressed with the examples, yours look better than the PS versions imo, and I love my PS XD

Yes, imo as well . In case of iWE each of the color channels contains high-frequency information registered by all the pixels. For example, in each row of the sensor only half are green pixels, but in my case the green channel contains the high-frequency information registered by all the pixels including the non-green pixels. So, in image #2 you can see the text information which is really hard (if possible at all) be extracted by demosaicing methods based on space-domain interpolation algorithms. The disadvantage of my method - it is very costly in terms of computing resorces.

Eye opening for sure!

Does it utilize lots of cores or lots of frequency?

Thanks for the extra details, love the work so far

Current version of iWE uses either CPU or GPU (graphics processing unit). In iWE you can ask for GPUs, and if they exist you can use one of them.

In case of CPU all the available cores are used in calculations. For example, my laptop has modern i7 with 8 cores working at up to 4.8 Ghz. The iWE algorithms are optimized and typically with this CPU it takes 10-15 second for applying 1 or 2 wave filters to 20 Mpx image. But in case of more wave filters are involved it can take up to one minute for the whole 20 Mpx image.

The GPU is multithread unit by definition. On my laptop I have two GPU units. The most powerfull has 16 multiprocessors with 1024 !!! threads per each multiprocessor. However, the GPU works at lower frequency (1.5 Ghz in my case). Also each time we ask GPU to do something, we must initialize it and pass a huge quantity of data to the GPU memory, which takes time. iWE uses floating point format, so single 20 Mpx image needs more than 300 M of memory. In my case using GPU allows getting of about 3X gain compared to my really powefull CPU. For example, to make the wave demosaicing of 20Mpx image it takes of about 15 seconds on my "Quadro T2000 with Max-Q Design" GPU.

 

[tpani]

Have you posted about this to the "Photographic Science & Technology"? It would be interesting to see what the people there comment on this.

 

[Serguei Palto]

Have you posted about this to the "Photographic Science & Technology"? It would be interesting to see what the people there comment on this.

I am not going to make a hype around iWE by participating at many forums simultaneously. I have enough knowledge in what I am doing, so I am not sure that extremely intensive forums discussions can be usefull. But, of course, everybody, who is interested in iWE, can join the discussion herein, and I will try to answer all the principal questions.

I post the information on iWE here, because I like and use m43, and hystorically the iWE project at very beginning aimed to enhance the signal/noise-output from m43 system. iWE, of course, supports almost all the existing systems (more than 1000 camera including the latest camera like OM-1). Everybody can use iWE . In fact, iWE is extremely simple in using, but, in any case, a user should spend some time to learn its basics (iWE is free, and learning is the only payment ).

By the way, I have made a short video on using enanced wave demosaicing with GPU. This video can be downloaded or viewed at the reference in my original post. The video is in English. Also at the same reference one can dowload the manual in English.

iWE exists, and that is why I am sharing it.

 

[Austinian]

Have you posted about this to the "Photographic Science & Technology"? It would be interesting to see what the people there comment on this.

I agree! I'd really like to see how this is viewed by some of the technical experts that participate in the PS&T forum.

 

[abgestumpft]

Hi Serguei,

nice upgrade
From my system the runtimes for 20MP OM-1 RAW with some spectral sharpening:
CPU (AMD 3700x) : 12,4sec
GPU (NVidia 1060 6GB): 9,2sec

Two things I recognized with iWE:
1. When there is some special character in File path (like: "ö") it won't load: moving the same image to another directory solves this (so minor issue, you just have to be aware about it)

2. I'm running iWE on a 32" 4K Screen. In Windows I have set scaling to 150%. Usually this scales only the Text/Icons etc. But in iWE this also seems to scale the Picture itself, so my 100% view is actually 150%. At least I guess that is what is happening, since in iWE the same image looks bigger than in other image applications (only Olympus Workspace seems to behave the same here ;-))

I found the friend of your bird here ;-)
OM-1 with Pana 100-300 II @300mm

100% crops:


Out of Camera JPG (cropped)


RAW edited in CaptureOne

Here my first Version in iWE:


RAW edited in iWE

I was continuing to edit the above Picture in iWE (removing the CAs) and then managed to crash the application (not sure what I exactly did at that time: I think I was moving around the preview Smart box to the very low bottom of the picture while playing with CA sliders).
Now when I load a RAW file (with LibRaw) it always shows a black/white picture with some fine dotted pattern -> one channel only? I removed all the iWE binaries and unzipped &/ run setup again, but still same issue. Can I somewhere cleanup the settings (of LibRaw)?

kind regards
Sebastian

 

[kalisti]

Wow, looks like you're making really good progress on this.

I'm very impressed with the examples, yours look better than the PS versions imo, and I love my PS XD

Yes, imo as well . In case of iWE each of the color channels contains high-frequency information registered by all the pixels. For example, in each row of the sensor only half are green pixels, but in my case the green channel contains the high-frequency information registered by all the pixels including the non-green pixels. So, in image #2 you can see the text information which is really hard (if possible at all) be extracted by demosaicing methods based on space-domain interpolation algorithms. The disadvantage of my method - it is very costly in terms of computing resorces.

Eye opening for sure!

Does it utilize lots of cores or lots of frequency?

Thanks for the extra details, love the work so far

Current version of iWE uses either CPU or GPU (graphics processing unit). In iWE you can ask for GPUs, and if they exist you can use one of them.

In case of CPU all the available cores are used in calculations. For example, my laptop has modern i7 with 8 cores working at up to 4.8 Ghz. The iWE algorithms are optimized and typically with this CPU it takes 10-15 second for applying 1 or 2 wave filters to 20 Mpx image. But in case of more wave filters are involved it can take up to one minute for the whole 20 Mpx image.

The GPU is multithread unit by definition. On my laptop I have two GPU units. The most powerfull has 16 multiprocessors with 1024 !!! threads per each multiprocessor. However, the GPU works at lower frequency (1.5 Ghz in my case). Also each time we ask GPU to do something, we must initialize it and pass a huge quantity of data to the GPU memory, which takes time. iWE uses floating point format, so single 20 Mpx image needs more than 300 M of memory. In my case using GPU allows getting of about 3X gain compared to my really powefull CPU. For example, to make the wave demosaicing of 20Mpx image it takes of about 15 seconds on my "Quadro T2000 with Max-Q Design" GPU.

I'm reading the manual first, and will install and give it a shot!

As a PS user I'm very curious to see how iWE works, and whether iWE is the complimentary tool to use beside PS (from your examples I think it is), darktable and such seeming too similar in function never found a place next to PS for me.

I'm excited to get started

Thanks for being so generous with your time!!

 

[ursamajorRO]

Cannot download it.

EDIT: I manage it. Nice work. I'll use it a bit to see what it can.

 

[Serguei Palto]

Hi Serguei,

nice upgrade
From my system the runtimes for 20MP OM-1 RAW with some spectral sharpening:
CPU (AMD 3700x) : 12,4sec
GPU (NVidia 1060 6GB): 9,2sec

Hi, Sebastian!

The GPU can provide larger gain in case of more wave filters are involved. Significant time is spent for initializing the GPU kernels, but if the job significant this time becomes less important.

Two things I recognized with iWE:
1. When there is some special character in File path (like: "ö") it won't load: moving the same image to another directory solves this (so minor issue, you just have to be aware about it)

Yes. I know this problem. It came with the latest LibRaw version, which exists for now only as the snapshot. I just downloaded the code from Github and compiled it as it is for using in iWE. This latest LibRaw version supports the OM-1, but has some localization problems - it does not support directories named with non-ASC symbols. Because I usually use only ASC symbols for naming the files (old tradition ) it has taken some time before I have realized this problem.

2. I'm running iWE on a 32" 4K Screen. In Windows I have set scaling to 150%. Usually this scales only the Text/Icons etc. But in iWE this also seems to scale the Picture itself, so my 100% view is actually 150%. At least I guess that is what is happening, since in iWE the same image looks bigger than in other image applications (only Olympus Workspace seems to behave the same here ;-))

Have in mind that first double clicking results in 1:1 view. The second double click will result to fit view, which in case of small images or large screen resolution can result in upscaling the image. Have also in mind that 1:1 view is very important for correct working the smart box mode.

I found the friend of your bird here ;-)
OM-1 with Pana 100-300 II @300mm

100% crops:


Out of Camera JPG (cropped)


RAW edited in CaptureOne

Here my first Version in iWE:


RAW edited in iWE

Nice bird!

iWe has a set of predifined denoise/sharpening filters on Detail panel. For efficient removing the background noise in low ISO images the N3 is quite usefull. You can also adjust second slider of "Deep Diff. Denoising" filter to suppress the backgroud noise without influencing the details.

I was continuing to edit the above Picture in iWE (removing the CAs) and then managed to crash the application (not sure what I exactly did at that time: I think I was moving around the preview Smart box to the very low bottom of the picture while playing with CA sliders).

For me it is very important to reproduce the crashing conditions, so I can look for the bug.

Now when I load a RAW file (with LibRaw) it always shows a black/white picture with some fine dotted pattern -> one channel only? I removed all the iWE binaries and unzipped &/ run setup again, but still same issue. Can I somewhere cleanup the settings (of LibRaw)?

Is it with "Enable Wave Demosaicing" switched OFF?

"Enable Wave Demosaicing" must be switched OFF in case you use LibRaw in standard mode.

"RAW Data" option must be unchecked on "Raw import and iWE Options/iWE RAW" panel.

What happens if you use "Free image Lib..." instead of LibRaw?

Please inform me on your success on fighting with the last problem.



Best regards, SP

kind regards
Sebastian

 

[abgestumpft]

Hi Serguei,

nice upgrade
From my system the runtimes for 20MP OM-1 RAW with some spectral sharpening:
CPU (AMD 3700x) : 12,4sec
GPU (NVidia 1060 6GB): 9,2sec

Hi, Sebastian!

The GPU can provide larger gain in case of more wave filters are involved. Significant time is spent for initializing the GPU kernels, but if the job significant this time becomes less important.

Two things I recognized with iWE:
1. When there is some special character in File path (like: "ö") it won't load: moving the same image to another directory solves this (so minor issue, you just have to be aware about it)

Yes. I know this problem. It came with the latest LibRaw version, which exists for now only as the snapshot. I just downloaded the code from Github and compiled it as it is for using in iWE. This latest LibRaw version supports the OM-1, but has some localization problems - it does not support directories named with non-ASC symbols. Because I usually use only ASC symbols for naming the files (old tradition ) it has taken some time before I have realized this problem.

2. I'm running iWE on a 32" 4K Screen. In Windows I have set scaling to 150%. Usually this scales only the Text/Icons etc. But in iWE this also seems to scale the Picture itself, so my 100% view is actually 150%. At least I guess that is what is happening, since in iWE the same image looks bigger than in other image applications (only Olympus Workspace seems to behave the same here ;-))

Have in mind that first double clicking results in 1:1 view. The second double click will result to fit view, which in case of small images or large screen resolution can result in upscaling the image. Have also in mind that 1:1 view is very important for correct working the smart box mode.

I found the friend of your bird here ;-)
OM-1 with Pana 100-300 II @300mm

100% crops:


Out of Camera JPG (cropped)


RAW edited in CaptureOne

Here my first Version in iWE:


RAW edited in iWE

Nice bird!

iWe has a set of predifined denoise/sharpening filters on Detail panel. For efficient removing the background noise in low ISO images the N3 is quite usefull. You can also adjust second slider of "Deep Diff. Denoising" filter to suppress the backgroud noise without influencing the details.

I will give it a try. Thanks for the tips!

I was continuing to edit the above Picture in iWE (removing the CAs) and then managed to crash the application (not sure what I exactly did at that time: I think I was moving around the preview Smart box to the very low bottom of the picture while playing with CA sliders).

For me it is very important to reproduce the crashing conditions, so I can look for the bug.

I will try to note down what I did when iWE should crash next time.
In Windows Eventlog I found this (not sure if that helps):

.NET Runtime ID: 1026
"Anwendung: Palto_Imaging.exe
Frameworkversion: v4.0.30319
Beschreibung: Der Prozess wurde aufgrund einer unbehandelten Ausnahme beendet.
Ausnahmeinformationen: ILGPU.Runtime.Cuda.CudaException
bei ILGPU.Runtime.Cuda.CudaKernel.Dispose(Boolean)
bei ILGPU.Util.DisposeBase.Finalize()"

Application Error ID: 1000
Name der fehlerhaften Anwendung: Palto_Imaging.exe, Version: 2.0.0.0, Zeitstempel: 0xa7da174a
Name des fehlerhaften Moduls: KERNELBASE.dll, Version: 10.0.19041.1645, Zeitstempel: 0x630193b4
Ausnahmecode: 0xe0434352
Fehleroffset: 0x0000000000034f69
ID des fehlerhaften Prozesses: 0x6954
Startzeit der fehlerhaften Anwendung: 0x01d850bf82f73ef7
Pfad der fehlerhaften Anwendung: C:\Users\Sebastian\AppData\Local\Apps\2.0\JAW191R9.DDT\MW301XJ8.54Q\palt..tion_0000000000000000_0003.0003_bb93156168f81985\Palto_Imaging.exe
Pfad des fehlerhaften Moduls: C:\WINDOWS\System32\KERNELBASE.dll
Berichtskennung: 1f6c572b-f9d9-4ff4-bb88-72ffbfc8176b
Vollständiger Name des fehlerhaften Pakets:
Anwendungs-ID, die relativ zum fehlerhaften Paket ist:

Now when I load a RAW file (with LibRaw) it always shows a black/white picture with some fine dotted pattern -> one channel only? I removed all the iWE binaries and unzipped &/ run setup again, but still same issue. Can I somewhere cleanup the settings (of LibRaw)?

Is it with "Enable Wave Demosaicing" switched OFF?

"Enable Wave Demosaicing" must be switched OFF in case you use LibRaw in standard mode.

"RAW Data" option must be unchecked on "Raw import and iWE Options/iWE RAW" panel.

This was it! Thanks! "RAW Data" was checked. When unchecking the RAW loads in full color again. I'm pretty sure I did not check it myself, but who knows... ;-)

What happens if you use "Free image Lib..." instead of LibRaw?

FreeImage... was still working fine (checked with EM1.3 files)

Please inform me on your success on fighting with the last problem.

Best regards, SP

kind regards
Sebastian

 

[Serguei Palto]

Hi Serguei,

nice upgrade
From my system the runtimes for 20MP OM-1 RAW with some spectral sharpening:
CPU (AMD 3700x) : 12,4sec
GPU (NVidia 1060 6GB): 9,2sec

Hi, Sebastian!

The GPU can provide larger gain in case of more wave filters are involved. Significant time is spent for initializing the GPU kernels, but if the job significant this time becomes less important.

Two things I recognized with iWE:
1. When there is some special character in File path (like: "ö") it won't load: moving the same image to another directory solves this (so minor issue, you just have to be aware about it)

Yes. I know this problem. It came with the latest LibRaw version, which exists for now only as the snapshot. I just downloaded the code from Github and compiled it as it is for using in iWE. This latest LibRaw version supports the OM-1, but has some localization problems - it does not support directories named with non-ASC symbols. Because I usually use only ASC symbols for naming the files (old tradition ) it has taken some time before I have realized this problem.

2. I'm running iWE on a 32" 4K Screen. In Windows I have set scaling to 150%. Usually this scales only the Text/Icons etc. But in iWE this also seems to scale the Picture itself, so my 100% view is actually 150%. At least I guess that is what is happening, since in iWE the same image looks bigger than in other image applications (only Olympus Workspace seems to behave the same here ;-))

Have in mind that first double clicking results in 1:1 view. The second double click will result to fit view, which in case of small images or large screen resolution can result in upscaling the image. Have also in mind that 1:1 view is very important for correct working the smart box mode.

I found the friend of your bird here ;-)
OM-1 with Pana 100-300 II @300mm

100% crops:


Out of Camera JPG (cropped)


RAW edited in CaptureOne

Here my first Version in iWE:


RAW edited in iWE

Nice bird!

iWe has a set of predifined denoise/sharpening filters on Detail panel. For efficient removing the background noise in low ISO images the N3 is quite usefull. You can also adjust second slider of "Deep Diff. Denoising" filter to suppress the backgroud noise without influencing the details.

I will give it a try. Thanks for the tips!

I was continuing to edit the above Picture in iWE (removing the CAs) and then managed to crash the application (not sure what I exactly did at that time: I think I was moving around the preview Smart box to the very low bottom of the picture while playing with CA sliders).

For me it is very important to reproduce the crashing conditions, so I can look for the bug.

I will try to note down what I did when iWE should crash next time.
In Windows Eventlog I found this (not sure if that helps):

.NET Runtime ID: 1026
"Anwendung: Palto_Imaging.exe
Frameworkversion: v4.0.30319
Beschreibung: Der Prozess wurde aufgrund einer unbehandelten Ausnahme beendet.
Ausnahmeinformationen: ILGPU.Runtime.Cuda.CudaException
bei ILGPU.Runtime.Cuda.CudaKernel.Dispose(Boolean)
bei ILGPU.Util.DisposeBase.Finalize()"

Application Error ID: 1000
Name der fehlerhaften Anwendung: Palto_Imaging.exe, Version: 2.0.0.0, Zeitstempel: 0xa7da174a
Name des fehlerhaften Moduls: KERNELBASE.dll, Version: 10.0.19041.1645, Zeitstempel: 0x630193b4
Ausnahmecode: 0xe0434352
Fehleroffset: 0x0000000000034f69
ID des fehlerhaften Prozesses: 0x6954
Startzeit der fehlerhaften Anwendung: 0x01d850bf82f73ef7
Pfad der fehlerhaften Anwendung: C:\Users\Sebastian\AppData\Local\Apps\2.0\JAW191R9.DDT\MW301XJ8.54Q\palt..tion_0000000000000000_0003.0003_bb93156168f81985\Palto_Imaging.exe
Pfad des fehlerhaften Moduls: C:\WINDOWS\System32\KERNELBASE.dll
Berichtskennung: 1f6c572b-f9d9-4ff4-bb88-72ffbfc8176b
Vollständiger Name des fehlerhaften Pakets:
Anwendungs-ID, die relativ zum fehlerhaften Paket ist:

Now when I load a RAW file (with LibRaw) it always shows a black/white picture with some fine dotted pattern -> one channel only? I removed all the iWE binaries and unzipped &/ run setup again, but still same issue. Can I somewhere cleanup the settings (of LibRaw)?

Is it with "Enable Wave Demosaicing" switched OFF?

"Enable Wave Demosaicing" must be switched OFF in case you use LibRaw in standard mode.

"RAW Data" option must be unchecked on "Raw import and iWE Options/iWE RAW" panel.

This was it! Thanks! "RAW Data" was checked. When unchecking the RAW loads in full color again. I'm pretty sure I did not check it myself, but who knows... ;-)

What happens if you use "Free image Lib..." instead of LibRaw?

FreeImage... was still working fine (checked with EM1.3 files)

Please inform me on your success on fighting with the last problem.

Best regards, SP

kind regards
Sebastian

OK. Fine!

Thanks for the error-report. At least I know where the problem has appeared, while the reason remains unclear.

 

[Ph0t0synthesis]

I'm a relative newcomer to this forum, and I think I need a bit of help.

I can see an improvement in the text image, but not in the photos of the great tit. Even without enlarging them, I can see a lot more detail in the Photoshop image (e.g. feather and bark texture), yet people are acclaiming the iWE version. To my eye, the latter looks washed out, but others seem to see a significant improvement.

Could someone enlighten me?

H

 

[Serguei Palto]

I'm a relative newcomer to this forum, and I think I need a bit of help.

I can see an improvement in the text image, but not in the photos of the great tit. Even without enlarging them, I can see a lot more detail in the Photoshop image (e.g. feather and bark texture), yet people are acclaiming the iWE version. To my eye, the latter looks washed out, but others seem to see a significant improvement.

Could someone enlighten me?

H

A text at controlled conditions is a very good object to see the demosaicing difference. I guess that what attracted your attention in PS image of the great tit - it is a bit higher local contrast, but the local contrast is not that I am talking about. The contrast can be adjusted in further processing to any tast. Also real life objects are bit difficult for estimation the resolution, because there are many factors which tend to decrease the resolution. Maybe in examples below you will be able to see the difference in resolution. The difference is in extremely fine feather details (the head of the tit is, of course, not the best part, because it is in continuous motion). And you must compare at 1:1 view (click original size and click again to go to the 1;1 view).

The first image (#1) is the output from the camera, while the second (#2) is from iWE at default settings (no sharpening or other filters) just after the wave demosaicing. The last one (#3) is from RAW converter of the Photoshop 2021.


#1. Out of camera photo


#2. iWE gpu-demosaicing default output (no sharpening).


#3 PS 2021 default output with no sharpening or denoising

 

[kalisti]

I'm a relative newcomer to this forum, and I think I need a bit of help.

I can see an improvement in the text image, but not in the photos of the great tit. Even without enlarging them, I can see a lot more detail in the Photoshop image (e.g. feather and bark texture), yet people are acclaiming the iWE version. To my eye, the latter looks washed out, but others seem to see a significant improvement.

Could someone enlighten me?

H

To my eye the detail is the similar, the PS version imo more clumsily represents the contrast.

People mistake detail for sharpening and adding contrast, neither of which adds detail, to me the iWE pic looks higher resolution, the detail looks finer.

I'm not a fan of sharpening for instance, to my eye always obvious and clumsy, even the new AI versions (which are certainly improved), but the threads and compliments run on for pages and pages, its just not my taste.

Just like HDR and images that people bring the shadow up in, that just have a medium brightness across the frame, also not my taste, some love it though.

I've quietly kept up to date with iWE, seeing someone doing work way over my head, but with good info and updates, this latest post impressed me enough to really want to try, and even though over my head the different way of processing interests me greatly.

I'm very curious to see how this processes skin and faces, and to import my results into PS afterwards to make finishing touches.

But its all down to (for me) that I like the difference between the PS and iWE image, perhaps not for all pictures all the time, but as mentioned in an earlier post, as a tool beside PS etc rather than instead.

Sorry about the mess of thoughts, I hope that helps.

 

[abgestumpft]

Hi Serguei,

nice upgrade
From my system the runtimes for 20MP OM-1 RAW with some spectral sharpening:
CPU (AMD 3700x) : 12,4sec
GPU (NVidia 1060 6GB): 9,2sec

Hi, Sebastian!

The GPU can provide larger gain in case of more wave filters are involved. Significant time is spent for initializing the GPU kernels, but if the job significant this time becomes less important.

Just an update on this: when I do a little bit more (Spectral Sharpening, Noise Reduction, CA Reduction), the improvement with GPU becomes bigger:
CPU (AMD 3700x) : 29,6sec
GPU (NVidia 1060 6GB): 11,6sec

2. I'm running iWE on a 32" 4K Screen. In Windows I have set scaling to 150%. Usually this scales only the Text/Icons etc. But in iWE this also seems to scale the Picture itself, so my 100% view is actually 150%. At least I guess that is what is happening, since in iWE the same image looks bigger than in other image applications (only Olympus Workspace seems to behave the same here ;-))

Have in mind that first double clicking results in 1:1 view. The second double click will result to fit view, which in case of small images or large screen resolution can result in upscaling the image. Have also in mind that 1:1 view is very important for correct working the smart box mode.

What I meant by this is:
In Windows I can change the scaling for Text, Icons etc. I use a 32" 4K Monitor and without this scaling the text and icons are really small. This is why I have set this to 150%

E.g. see here: https://www.eizoglobal.com/support/compatibility/dpi_scaling_settings_windows/

In iWE this setting also has an effect on the zoom level of the image itself, not only the GUI/Icons/Text.

This is how 100% zoom looks like (Left: Affinity Photo 100% - Right: iWE 100%) when Scaling is set to 150% in Windows:


(Left: Affinity Photo 100% - Right: iWE 100%) - Windows scaling 150%

When I change the scaling in Windows to 100% it looks like this:


(Left: Affinity Photo 100% - Right: iWE 100%) - Windows scaling 100%

I was continuing to edit the above Picture in iWE (removing the CAs) and then managed to crash the application (not sure what I exactly did at that time: I think I was moving around the preview Smart box to the very low bottom of the picture while playing with CA sliders).

For me it is very important to reproduce the crashing conditions, so I can look for the bug.

Since the last crash I did not have another one -> running stable

 

[Serguei Palto]

Hi Serguei,

nice upgrade
From my system the runtimes for 20MP OM-1 RAW with some spectral sharpening:
CPU (AMD 3700x) : 12,4sec
GPU (NVidia 1060 6GB): 9,2sec

Hi, Sebastian!

The GPU can provide larger gain in case of more wave filters are involved. Significant time is spent for initializing the GPU kernels, but if the job significant this time becomes less important.

Just an update on this: when I do a little bit more (Spectral Sharpening, Noise Reduction, CA Reduction), the improvement with GPU becomes bigger:
CPU (AMD 3700x) : 29,6sec
GPU (NVidia 1060 6GB): 11,6sec

2. I'm running iWE on a 32" 4K Screen. In Windows I have set scaling to 150%. Usually this scales only the Text/Icons etc. But in iWE this also seems to scale the Picture itself, so my 100% view is actually 150%. At least I guess that is what is happening, since in iWE the same image looks bigger than in other image applications (only Olympus Workspace seems to behave the same here ;-))

Have in mind that first double clicking results in 1:1 view. The second double click will result to fit view, which in case of small images or large screen resolution can result in upscaling the image. Have also in mind that 1:1 view is very important for correct working the smart box mode.

What I meant by this is:
In Windows I can change the scaling for Text, Icons etc. I use a 32" 4K Monitor and without this scaling the text and icons are really small. This is why I have set this to 150%

E.g. see here: https://www.eizoglobal.com/support/compatibility/dpi_scaling_settings_windows/

In iWE this setting also has an effect on the zoom level of the image itself, not only the GUI/Icons/Text.

This is how 100% zoom looks like (Left: Affinity Photo 100% - Right: iWE 100%) when Scaling is set to 150% in Windows:


(Left: Affinity Photo 100% - Right: iWE 100%) - Windows scaling 150%

When I change the scaling in Windows to 100% it looks like this:


(Left: Affinity Photo 100% - Right: iWE 100%) - Windows scaling 100%

I was continuing to edit the above Picture in iWE (removing the CAs) and then managed to crash the application (not sure what I exactly did at that time: I think I was moving around the preview Smart box to the very low bottom of the picture while playing with CA sliders).

For me it is very important to reproduce the crashing conditions, so I can look for the bug.

Since the last crash I did not have another one -> running stable

Hi, Sebastian,

Thanks for the good info in general.

I think that the "scaling behavior", which is different from Affinity one, comes from MS Windows graphics library I use, where the objects want to be "clever" It seems, they do the hidden "useful job" taking into account global System settings. I do not like this, and hope that this does not influence the iWE performance.

At the reference I have mentioned in the original post a new (10-th) edition of iWE can be downloaded. Therein I have made further optimization of the wave-GPU-demosaicing itself and some filters related to sharpening and denoising at the demosaicing stage. These filters are OFF by default, but they can be turned ON with checking options on the RAW panel.