Topaz just released DeNoise AI

[piggsy]

That's strange as I had no trouble processing tiff's in AI Denoise. I convert to Tiff in Oly Workspace and do all my processing on the Tiff's. I'm not using the trail version but it shouldn't make any difference. I'm using it as a plugin in Paintshop Pro.

Not one single file I have on my entire computer I've tried so far, exported by either lightroom or photoshop, can be read. If I go in and save it from irfanview as an uncompressed tiff, it'll read em. Couldn't be more bog standard (srgb, 16b, zip). Gigapixel sitting right next to it loads the same files just fine. Pretty funny.

 

[piggsy]

I contacted Topaz because of the long time it takes to “Apply” (save) Denoise AI. They said five minutes or more to do this for JUST one image is within normal limits. The last two I tried to save took 15-20 minutes. They suggested an upgraded GPU (my computer has lots of RAM & memory and up to date graphics driver) on my only two year old computer. I got this and Sharpen AI (same issues) as upgrades so didn’t lose any $$ but it does seem that they should be developing stuff like this for a wider range of computers. Just IMHO.

If you skip down to "difference between cpu and gpu" here it has a reasonable explanation. It's a different style of computation that the gpu hardware is more suited towards performing.

Why are GPUs necessary for training Deep Learning models?

 

[Lawrence Beck]

"In any event, using the term crushes is a bit hyperbolic."

Jerry, I'm in the process of cleaning the egg off my face. I made the mistake in presuming that since DeNoise AI did a better job on a couple images that I tested it was therefore best. I've since learned that there are shortcomings that need to be addressed... so I thank you for the comment you made about "hyperbole" as it has caused me to dig a bit deeper. You were spot on.
While I'm confident that Topaz will address isues that some will hopefully bring to their attention, I'm also aware that the competition (Noiseware, DXO Prime etc) will also be working overtime to improve their product. Product survival is what's at stake here, and companies who ignore this will suffer the inevitable consequences.
An issue that concerns me is the way DeNoise AI smears detail in shadows. I've been searching for reviews on DeNoise AI to see what others have found and HDR shooter found an issue that I've also seen in several tests. Link here: Topaz Denoise AI from Topaz Labs - HDRshooter
Look at the first test of the three columns. First is the original noisy image followed by a Noiseware image and finally DeNoise AI.
While the part of the columns in direct light looks best on the Topaz sample (to my way of thinking), look into the shadows just right of center, and note that DeNoise AI has smeared the texture in the shadows. I think that the detail that's lost in the face of the columns in the Noiseware sample can be restored with judicious use of a sharpening program, while the detail that's smeared in the shadows of the DeNoise AI sample are forever lost.
I'm going to send Topaz a link to this test, as this is something that needs to be examined.
Lesson to be learned here: each of us have different concepts as to what makes our images look best. A few tests showing superior results with one piece of software does not a conclusion make. Know that this is a highly competitive market, and those who are driven to winning the race are always trying to improve their products. Apart from my underwater photography I always shoot at low iso as image quality is of paramount importance to me. But a good noise reduction program will allow for images taken underwater at the fastest flash sync, smaller fstops used for better depth of field (for macro), and lowest iso... so noise reduction is an area that I'll continue to explore as our m43 sensors need all the help they can get at high iso.

 

[JLGF1]

What I've experienced so far is that for some images AI-DN works very well and for others not so much. I guess that is to be expected in the early stages of the AI training. I, and others, have had AIDN produce some minor weird artifacts and some major weird issues. The latter they have either fixed or are investigating. The more (bad) examples we give them the quicker the product will improve, hopefully. For example, Canon Raws (or maybe any) with initial high color noise are not handled as well as some other NR apps. Perhaps AIDN hasn't been adequately trained for that, yet.

This was one of the major issues, but there are work-arounds:

Subscribe to see EXIF info for this image (if available)

 

[runner girl]

That is what the app's "Help->Graphics Info" menu tells you...

Intel HD Graphics 620. I did update the firmware.

 

[JLGF1]

Intel HD Graphics 620. I did update the firmware.

That is better than mine. I don't think processing time will ever fall below one minute for us unless we get a faster (dedicated) GPU with 4GB+ of video memory. Make sure to set your allowed video memory to High in the app prefs.

Subscribe to see EXIF info for this image (if available)

 

[runner girl]

That is better than mine. I don't think processing time will ever fall below one minute for us unless we get a faster (dedicated) GPU with 4GB of video memory. Make sure to set your allowed video memory to High in the app prefs.

Thx I will check that. I don’t mind a few minutes, but 20 minutes is beyond excessive.

 

[JamesD172]

Intel HD Graphics 620. I did update the firmware.

There's the issue; you're using an integrated - not dedicated - GPU. If you plan to use your computer for GPU-intensive tasks, e.g. video encoding, machine learning (or AI applications such as this), or gaming, you'll need a dedicated graphics card. They don't have to be uber expensive either. A 1050 Ti would probably give you a huge boost in speed, and it's $169.

Edit: Another good option would be the newly released Nvidia 1650, which performs similarly to the above but only costs $149

 

[JLGF1]

Ran this test image through AIDN and got some nice results.

Subscribe to see EXIF info for this image (if available)

Before/Original

Subscribe to see EXIF info for this image (if available)

After AIDN

Subscribe to see EXIF info for this image (if available)

 

[JLGF1]

Interesting new thread here...some users are deploying DxO-Prime and then AIDN. Raw test image available for d/l.

Another DeNoise vs DxO PRIME thread: Retouching Forum: Digital Photography Review

 

[runner girl]

There's the issue; you're using an integrated - not dedicated - GPU. If you plan to use your computer for GPU-intensive tasks, e.g. video encoding, machine learning (or AI applications such as this), or gaming, you'll need a dedicated graphics card. They don't have to be uber expensive either. A 1050 Ti would probably give you a huge boost in speed, and it's $169.

Edit: Another good option would be the newly released Nvidia 1650, which performs similarly to the above but only costs $149

Unfortunately I can’t change it out on my laptop so I’ll be happy with AI Clear.

 

[betamax]

Interesting new thread here...some users are deploying DxO-Prime and then AIDN.

I did some testing over the last couple of days and I too found, I think based on a comment on this thread, that removing some of the chroma nosie prior to treating with Dnoise, in my case using Capture 1, reduces artifacts quiet a bit and results in a pretty good image.

I do like how denoise also sharpens. It's a type of sharpenning I like but haven't been able to replicate myself. It's probably this that's swaying me at the moment to get it.

 

[JLGF1]

Samples: DeNoise vs DxO PRIME vs Prime plus DeNoise: Retouching Forum: Digital Photography Review

 

[JLGF1]

More comparisons with DxO-Prime:

DeNoise AI - Comparison

 

[betamax]

I found under Windows 10, under help > graphics info it was still showing the Intel 620 graphics. I had to manually set the NVIDIA Control Panel to use the GPU - a MX150 - for Denoise AI. And now it's a good deal faster for me taking around 40 seconds instead of nearly 140 seconds. It shoud have switched automatically to use the GPU, but.... Windows 10 insider program version.

 

[Drdave944]

I have not used Topaz before but was getting good results on trial . The item was on sale for$59 so I went for the trial . When I decided to buy ,the price was $79.
I hate the feeling of bait and switch tactics like this. It could have been a beautiful friendship.

 

[JLGF1]

All sales have end dates and the trials do not have price lock-in, so that is hardly bait-switch. If you email sales I'd say there is a good chance they might still give you the sale price, but it's really your failure for not keeping up with it.

 

[Drdave944]

All sales have end dates and the trials do not have price lock-in, so that is hardly bait-switch. If you email sales I'd say there is a good chance they might still give you the sale price, but it's really your failure for not keeping up with it.

I don't know what business you are in,but if you operate that way,you could improve customer satisfaction by not appearing to practice bait and switch tactics and holding customers to the fine print.

 

[gwydionjhr]

I don't know what business you are in,but if you operate that way,you could improve customer satisfaction by not appearing to practice bait and switch tactics and holding customers to the fine print.

I don't think he's off base. The sale had a deadline and you waited too long.

Have you looked around for a promo code? Called support to ask for the introductory price as he suggested?

FWIW, I've run dozens of pictures through DN AI, and I'm super happy with it.

 

[JLGF1]

You missed this one, too, but they have them often for those that stay on top of things.

Subscribe to see EXIF info for this image (if available)