DeNoise 3.5.0

For anyone wondering, there is still no RAW support for Sony A1 files in this update - is there a date for this yet?

1 Like

I noticed when I drag & drop a RAW photo into the standalone DeNoise AI 3.5 window, the RAW mode is greyed out. Apparently it does not recognize the photo as RAW. I have tried this with Canon CR2 Raw images. Am I missing something here? Thank you for any help.

Hi kd -

No issue here with Canon CR2 and CR3 RAW files.

Win10 Pro, 21H2 build 19044.1469
AMD Ryzen 7 / 16GB RAM / AMD R9 200 series video

Just tried with macOS 12.2 and still fails, I take it you noticed my logs in DeNoise 3.5.0 - #19 by vargolsoft

I have this issue in 3.4.2 and 3.5. M1 Max MacBook Monterrey 12.2

Still no support for a camera that came out almost a year ago? That doesn’t really give me much hope for Z9 support any time soon.

Possible Memory Leak Processing Raw Files

I have been running a lot of DeNoise AI tests across different computers with different storage speeds and different raw file types. My objective was to get information on what hardware combinations provides the fastest processing times in DeNoise AI and share that on my Photo Taco Podcast.

In the process I have observed what looks to me like a memory leak when DeNoise AI is processing raw files. I first noticed significant swap file usage on M1 Macs with less than 64GB of RAM as they process raw files. For ARW, CR2, CR3, and RAF raw types the swap usage is pretty extreme, but I could process 300 of those raw files. However, I have found that I can’t process 300 NEF raw files from a Nikon D850 without DeNoise AI failing.

Here is the testing scenario:

  1. Open 300 Nikon D850 NEF files in DeNoise AI 3.4.2 or 3.5
  2. Set them to be processed as Severe Noise
  3. Set them to be processed with Settings of Auto
  4. Set the Auto Remove Noise to 15
  5. Set the Auto Enhance Sharpness to 60
  6. Set the Post-Processing Recover Original Detail to 50
  7. Set the Post-Processing Color Noise Reduction to 10
  8. Set the AI-Processor to be the GPU in DeNoise AI Preferences
  9. Set the Allowed Memory Consumption to be High (though the behavior seems to be the same with medium and low) in DeNoise AI Preferences
  10. Make sure all 300 images are selected to process and hit the Save Images button
  11. Select TIF for image format
  12. Select ZIP for compression
  13. Select 16 for bit depth
  14. Select Auto for filename
  15. Select Preserve source profile for Color profile
  16. Choose anywhere for the save directory
  17. Hit the Start button

Before running the test I reboot the computer and shut down any applications that launch on startup. I am recording metrics about how hardware is being utilized and I didn’t want to have those metrics influenced by anything else that might be running.

On M1 Macs (13" M1 original and 14" M1 Pro) with 16GB of RAM the swap usage while processing raw files constantly grows to between 15 and 46GB of swap space even though the RAM on the computer doesn’t get used above 11GB. For all but the Nikon D850 NEF files there is some recovery after reaching that peak of swap space where 25% of the swap is released during the processing at which point the swap begins to grow again. There is a sawtooth pattern in the swap usage from that point growing back to the peak pretty quickly and releasing 25%.

The behavior with the Nikon D850 NEF files is a little different. In that case the swap grows until it reaches 50GB and then MacOS stops DeNoise AI from processing and shows an error message saying “Your system has run out of application memory.” This happens at about image 125 to 130 on both MacOS Big Sur 11.6.3 and MacOS Monteray 12.2. DeNoise AI doesn’t release any swap with these files.

Here are the things I have tried to workaround this issue where I can’t process the 300 NEF images from the Nikon D850:

  1. Process the images 100 at a time. This was successful. All 300 images processed, but I had to close DeNoise AI between each 100 image run so that the memory and swap it was using would get released.
  2. Convert NEF to DNG using Adobe DNG Converter. The processing failed exactly the same way as the raw NEF.
  3. Convert NEF to TIF using Photoshop. This was successful. Even better, the swap file wasn’t used at all here! Downside is conversion to TIF takes nearly as long as the processing DeNoise AI has to do on them.
  4. Process these same NEF files on a PC with Windows 10 19042 32GB of RAM and a GTX 1060 3GB video card. This also failed. It failed faster than on the M1 Macs. It failed between 52 and 57 of these images. The error message shown in DeNoise was cut off so this is what I could see:

    “Error: OpenCV(4.5.3) C:\Users\Adam.conan\data\opencv\4.5.3\
\
\build/eae637997668fbc59b3bd2e17a93770838ebb99b2\aource_subfolder\modules\core\src\alloc.cpp:73 error: (-4 insufficient m”.

    BTW, my computer doesn’t have an “Adam” user so “C:\Users\Adam” isn’t a valid directory.

    This was also in the Windows 10 Event Logs:

    “Windows successfully diagnosed a low virtual memory condition. The following programs consumed the most virtual memory: Topaz DeNoise AI.exe (14012) consumed 30708682752 bytes, vmmem (4360) consumed 631721984 bytes, and MsMpEng.exe (6724) consumed 308805632 bytes.”
  5. Process the images 50 at a time on the PC. This was successful. All 300 images processed but like on the Mac I had to close DeNoise AI between each run so that memory could be released.
  6. Process NEF files from Nikon D3400. This was successful. 300 images were processed on all machines. The swap was heavily used, but at least this followed the same pattern as the other raw file types with 25% of the swap being released before it built back up so that it stayed under 50GB.

I haven’t tried running this process on a 16" M1 Max with 64GB of RAM yet, I am having a friend to that soon. I did have that friend run this batch on the CR3 files from a Canon R6 and it was able to successfully process 300 CR3 files from a Canon R6 without using swap, but DeNoise AI claimed 40GB of RAM to do so.

It looks to me like there is a fairly significant memory leak issue in DeNoise AI related to raw processing on both Mac and Windows leading to significant swap file use, and the swap usage being particularly bad for NEF files from a Nikon D850.

1 Like

Curious if anyone has noticed DeNoise AI introducing changes in colors/lighting, etc.?

I recently encountered this for the first time after using DeNoise AI for a couple of years without such issues.

I read recently that DXO DeepPrime on the Mac M1 gives a choice of using the CPU, GPU, or Neural Engine and that the Neural Engine is much faster than using the GPU. Does Denoise AI give those choices also on the M1?

No directly, but I’ve just render a slideshow using ffmpeg for some images processed in Denoise and Sharpen and the colours are every flat in the video compared to the images themselves. One image of a sand stone building with an overcast sky looked practically grey scale.

Thats the first time thats happened.

The only difference I can tell from old images is that the Color Sync Profile has changed from P3 to ProPhoto

Oh and it seems for 3.5.0 the first image rendered being a pixel short in both dimensions is back, at least for the RAW model.

I saw several mentions that DXO 5 gives the choice on M1 of using the CPU, GPU, or Neural Engine and that the Neural Engine is fastest. Here are just 2 things I saw:

And here is another one:

Not yet. I’ve had a support ticket open on this for a while. Got a response a few days ago saying that the latest version of Gigapixel is fixed and they are working on the fix for Denoise. The latest version of Sharpen works fine as well.

Several times my Photoshop opens an instance of Topaz, the notification popup shows up.
I am able to go into Topaz, but i think another window has openend (which I cannot seem to activate). Nothing is responsive anymore.

I need to Force Quit Topaz and Photoshop (and all my work is not saved) :-/

Same here lost half a days work every time using Sharpen Ai a notice for update comes up clicking on update does not work and computer crashes-- not a happy camper

Hello,

premise, I still use Win 7 Ultimate 64 bit, and since the last two updates DeNoise does NOT work anymore, it keeps giving me an error at startup.

The version that works for me is my current V3.4.4.

Any solution ???

Let me know and thanks.
Cheers,
Michele

Hello,

I have not heard of this issue yet at least for this specific file type although I know some lens data gets messed up in some other formats. I see you can’t upload files here directly, so I created a dropbox folder that you should be able to upload to. It should have been e-mailed to the address you use for this forum.

Do you mean 3.3.4? 3.4.0 was the first version where we officially moved to Qt 6 in order to support newer operating systems. Qt has officially dropped support for Windows 7 and 8, although it seems some users are still able to get it running and I believe it runs on our test machine.

If you give more info about the error there may be a solution, but if it’s related to Qt itself then we cannot revert back since it will break forward compatibility.

Thanks for the response Adam. I have not yet received the DropBox link, and it is not in my junk mail etc.If you like I could put them in a publicly accessible read-only folder on my OneDrive and post here?

If you’re fine with them being public then that works too. If not I can also send you a link via a direct message on here instead. This is the first time I’ve tried the e-mail option on dropbox but it seems like it doesn’t work.

No problem. I have put together a folder with relevant images, you can access the web folder via this link : https://1drv.ms/u/s!Amn7hEj1uk28gbpUgPNzC2GhSiQF0A?e=0MJRPf

The link will expire in 2 weeks and the access password is : DeNoiseAccessFeb2022

The files are:

#1- The RAW file direct from the Panasonic camera in .RW2 extension

#2 - The file processed through DeNoiseAI in Auto mode and output with .DNG extension

#3 - A JPG conversion of the RAW file from Lightroom

#4 - A JPG conversion of the DeNoiseAI processed .DNG from Lightroom. If you flick back and forward between #3 and #4 you can immediately see the missing lens correction data from the DeNoise DNG creates issues such as vignetting and barrel distortion, visible on the lines of the drawers in the background.

#5 - A screen snip of the Camera & Lens info section of Lightroom’s Info panel, for the original RAW

#6 - The same area, but for the DeNoise processed .DNG. Note the completely corrupted Lens ID, although for some reason the Camera ID is retained.

#7 - Shows how, with the lens correction profile left embedded, Lightroom recognises it and automatically corrects the lens. This is from the original RAW.

#8 - As this information is not retained by the DeNoiseAI DNG, there is no auto correction, and Micro Four Thirds lenses are not available in the manually selectable options if you tick Enable Lens Correction as Lightroom assumes that for M4/3 lenses the correction data will already be embedded.

Hope this helps :slight_smile:

2 Likes