IMPORTANT UPDATE: You can open the Topaz Ai files in Darktable if you do this first th-cam.com/video/0dpVqydHeFU/w-d-xo.html Topaz DeNoise Ai 3.7.2 on Intel macs does not have the problem!
Hi Andy. Thanks for another informative video. Just curious, after you've used DxO or Topaz Denoise to process a raw file (and thus the demozaicing process is done), is there any real benefit to saving the output as a DNG over a 16-bit TIFF, given that the image information in the DNG file is going to in RGB format anyway ?
Hi Stephen - there's just one benefit that I can see, and that's you can still process it within a linear working space, which you would not be able to do if it was a 16bit TIFF.
The AI denoising really is amazing. I almost feel cheated (or maybe that's not the right word, but anyway, I'm tired) because I learned the wavelet denoising techniques of Raw Therapy (thanks to your excellent videos, Andy) and spent hours applying them. Then I discovered topaz a month later... Not to say it was a waste of time, but man there is no competition. Truly stunning what they can do. In the future I see camera makers baking this stuff into the camera. Flip the setting on and no more ISO noise ever (well, within reason, but who knows how far the AI stuff will come).
Hi Phillip, yes, it is amazing - and in my mind, people ought to take it all into consideration before buying a new camera, especially if they are wildlife shooters. Personally, I doubt it would ever find it's way into an actual camera though. The processing power requirements would be huge - remember, the user might be shooting at 14 frames per second or more!
Very informative video, as always, thank you Andy. Having seen all of your RT Videos I had to try the challenge. Of course I failed, did not think that I would be able to come close. How should I, having learned all RT stuff from you ;) However I was also curious about how AI based denoising would compare to my RT skills on my own photos. So I downloaded the trial versions of Topaz Denoise AI and DxO PureRaw (actual versions, just 3 days ago). Had to boot my second computer to the windows partition that I hardly ever use but keep just for these kind of instances. I am on linux for years so LR is no option. By the way I am quite happy with RT and Gimp... Quite impressed with the files from you, my own ones where of poorer quality for a few reasons (MFT Sensor, birds smaller in the frame, more noise..). My RT attempts with Local Adjustment Denoise, the standard one in the details tab as well as with wavelets and also combinations did not come close and it shoud not surprise you. But with my own files I noticed some heavy artifacts with topaz as well as rendered detail that is absent in the original file which is really quite strange. Since Denoise AI has more control over parameters I tried to change settings and played around with them. Used their different algorithms but ended up with the following result: For my Photos (Olympus E-M1 III and Panasonic G9) DxO PureRaw had consistently better output (best with the DeepPrimeXD which on my old machine took 15 min for each 20 mpx file :( ) than Topaz Denoise AI (which btw was much faster at about 6 min per file). Still considering purchasing DxO for the reasons you give at the end of the video but fear that you cannot do much tweaking if the result does not satisfy you (I will continue testing). But have you encountered (with other images) strange artifacts or halos or details that are not only oversharpened but one can only call "creative"? If I look at the images from you that I downloaded from the link both apps seem to work pretty well. Sorry for the long comment. Kind regards
Cheers for that Zweima! Topaz Denoise Ai and Photo Ai can both produced some strange artefacts if you are not careful, and they are very hard to near impossible to see in the 100% previews. DXO PR3 - for me at least - doesn't produce any artefacts as such, but if you compensate for lens blur then I find its sharpening a little aggressive. Without a doubt, though, I find it better than the Topaz variants on the whole. And if you ever tried DarkTable, it has problems reading the Topaz dng files, but the DXO dngs can be read by DT no problem. If you want to send me a couple of your own files, so I can see where the differences are etc, then feel free - use WeTransfer to tuition@wildlifeinpixels.net
Hi. Looking at those 4 shots you sent, the Grebe, the Peregrine, and what I presume MIGHT be some type of Bee-eater, are badly out of focus. The reason noise is more apparent in out of focus areas is that noise 'happens' on the sensor, and is therefore always razor sharp. However, the fourth shot - is that a female Redstart? - is tack sharp, and so what noise is in the image is pretty much invisible in the bird itself, and only visible in the background. Let's look at your Grebe shot. Download this JPEG from my Dropbox www.dropbox.com/s/vd119w3m3i8l3s6/P1080487-RW2_DxO_DeepPRIMEXD-Edit-2.jpg?dl=0 I took the raw file and ran it through DXO Deep Prime 3 with NO SHARPENING or other corrections EXCEPT Chromatic Aberration. The resulting dng file was then sent to Topaz Sharpen Ai (not PhotoAi) with a small amount of MOTION BLUR Normal correction! Considering how noisy and out of focus the original is, I don't think that's too bad, and no raw processing to speak of, except that done by DXO in the first place. What do you think?
Hello Andy, wow and thank you so much for taking the time. (Strangely my answer yesterday was not posted?! )As you write, the result looks pretty good considering the original. I tried myself to sharpen and optimize via RT from the PR3 Output and am very pleased. That should be the way for me to go. The supposed Redstart (don't know what it is TBH) is of much better quality mainly because of the gathered light. My own editing on that one without using AI tech is quite satisfactory to me. Don't know, if you had a look at the pp3. I am not sure if the pics are out of focus. I feel that it is more the low light (Shutter speed and available light) in combination with the small sensor (more prone to noise) and blurring by heat haze (which may also throw off the focus). The "Redstart" is F/6.3 1/320s ISO400, while the Peregrine (I think it is a hobby?/falco subbuteo) is F/9 1/1600s ISO 5000 and it is still underexposed cause it was getting dark already. But that was a pretty warm and sunny day so heat haze might very likely have been an issue. The grebe is an old shot from about 10 years ago with a much worse lens/camera combo and I made it from my canoe...
On Linux you can run MacOS or Windows on a KVM based virtual machine. If you got the hardware you can even pass a GPU to the virtual machine and run the VM at near native speeds as if it were running native on the machine. A lot of gamers are run windows on linux with GPU pass through and getting just as good performance, but even if you don't need such power you can still get a basic VM setup to do simple task such as denoise. Another alternative would be to see if one can get the apps running with WINE or Play on Linux but not all apps run well with those emulators.
That old non-Ai version does not do the same thing as either Denoise Ai or Photo Ai Juan, so you are missing out! The Darktable devs have just had a sample from me to investigate, but have you seen my supplementary video, linked in the pinned comment?
@@AndyAstbury Yeah, that's why I pointed out that has to be something related to the new versions ( I don't have one to check it out myself). I've only tried with the last one I've got.
@@Juan-wu9vb Non-Ai Topaz DeNoise does not do Raw Denoising. If you are on Windows or Intel Mac (not M1/Apple Silicon), getting the DeNoise Ai version 3.7.2 will blow your mind!
Interesting comparison. You can run the output DNG from Topaz through the Adobe DNG converter, and it will load in Darktable without any sculls (for some obscure reason) - still the demosaicing is a done deal. The further editing seems to be as smooth as with the DxO DNGs (I only use the latter in Darktable). You can zoom with the scroll wheel. Another pleasant tool in Darktable for those who want to use DxO without any sharpening - is the Diffuse or sharpen module. It can even soften sharpening in specific ways and areas. It's not easy to grasp right away, but Boris Hajdukovic has made the following video, which imo is the best to explain this specific tool (observe the ability to blur an edge, while keeping contrast around it): th-cam.com/video/pAbyORw0mng/w-d-xo.html
IMPORTANT UPDATE: You can open the Topaz Ai files in Darktable if you do this first th-cam.com/video/0dpVqydHeFU/w-d-xo.html
Topaz DeNoise Ai 3.7.2 on Intel macs does not have the problem!
Hi Andy. Thanks for another informative video. Just curious, after you've used DxO or Topaz Denoise to process a raw file (and thus the demozaicing process is done), is there any real benefit to saving the output as a DNG over a 16-bit TIFF, given that the image information in the DNG file is going to in RGB format anyway ?
Hi Stephen - there's just one benefit that I can see, and that's you can still process it within a linear working space, which you would not be able to do if it was a 16bit TIFF.
@@AndyAstbury Ah, that makes sense. Thanks.
The AI denoising really is amazing. I almost feel cheated (or maybe that's not the right word, but anyway, I'm tired) because I learned the wavelet denoising techniques of Raw Therapy (thanks to your excellent videos, Andy) and spent hours applying them. Then I discovered topaz a month later... Not to say it was a waste of time, but man there is no competition. Truly stunning what they can do.
In the future I see camera makers baking this stuff into the camera. Flip the setting on and no more ISO noise ever (well, within reason, but who knows how far the AI stuff will come).
Hi Phillip, yes, it is amazing - and in my mind, people ought to take it all into consideration before buying a new camera, especially if they are wildlife shooters.
Personally, I doubt it would ever find it's way into an actual camera though. The processing power requirements would be huge - remember, the user might be shooting at 14 frames per second or more!
Very informative video, as always, thank you Andy. Having seen all of your RT Videos I had to try the challenge. Of course I failed, did not think that I would be able to come close. How should I, having learned all RT stuff from you ;) However I was also curious about how AI based denoising would compare to my RT skills on my own photos. So I downloaded the trial versions of Topaz Denoise AI and DxO PureRaw (actual versions, just 3 days ago). Had to boot my second computer to the windows partition that I hardly ever use but keep just for these kind of instances. I am on linux for years so LR is no option. By the way I am quite happy with RT and Gimp... Quite impressed with the files from you, my own ones where of poorer quality for a few reasons (MFT Sensor, birds smaller in the frame, more noise..). My RT attempts with Local Adjustment Denoise, the standard one in the details tab as well as with wavelets and also combinations did not come close and it shoud not surprise you. But with my own files I noticed some heavy artifacts with topaz as well as rendered detail that is absent in the original file which is really quite strange. Since Denoise AI has more control over parameters I tried to change settings and played around with them. Used their different algorithms but ended up with the following result: For my Photos (Olympus E-M1 III and Panasonic G9) DxO PureRaw had consistently better output (best with the DeepPrimeXD which on my old machine took 15 min for each 20 mpx file :( ) than Topaz Denoise AI (which btw was much faster at about 6 min per file). Still considering purchasing DxO for the reasons you give at the end of the video but fear that you cannot do much tweaking if the result does not satisfy you (I will continue testing). But have you encountered (with other images) strange artifacts or halos or details that are not only oversharpened but one can only call "creative"? If I look at the images from you that I downloaded from the link both apps seem to work pretty well. Sorry for the long comment. Kind regards
Cheers for that Zweima!
Topaz Denoise Ai and Photo Ai can both produced some strange artefacts if you are not careful, and they are very hard to near impossible to see in the 100% previews.
DXO PR3 - for me at least - doesn't produce any artefacts as such, but if you compensate for lens blur then I find its sharpening a little aggressive.
Without a doubt, though, I find it better than the Topaz variants on the whole.
And if you ever tried DarkTable, it has problems reading the Topaz dng files, but the DXO dngs can be read by DT no problem.
If you want to send me a couple of your own files, so I can see where the differences are etc, then feel free - use WeTransfer to tuition@wildlifeinpixels.net
@@AndyAstbury Hello Andy. I just sent you a few pics. Hope you got them. Thanks alot
Hi. Looking at those 4 shots you sent, the Grebe, the Peregrine, and what I presume MIGHT be some type of Bee-eater, are badly out of focus.
The reason noise is more apparent in out of focus areas is that noise 'happens' on the sensor, and is therefore always razor sharp.
However, the fourth shot - is that a female Redstart? - is tack sharp, and so what noise is in the image is pretty much invisible in the bird itself, and only visible in the background.
Let's look at your Grebe shot.
Download this JPEG from my Dropbox www.dropbox.com/s/vd119w3m3i8l3s6/P1080487-RW2_DxO_DeepPRIMEXD-Edit-2.jpg?dl=0
I took the raw file and ran it through DXO Deep Prime 3 with NO SHARPENING or other corrections EXCEPT Chromatic Aberration.
The resulting dng file was then sent to Topaz Sharpen Ai (not PhotoAi) with a small amount of MOTION BLUR Normal correction!
Considering how noisy and out of focus the original is, I don't think that's too bad, and no raw processing to speak of, except that done by DXO in the first place.
What do you think?
@@AndyAstbury Hi Andy. Strange thing is, my replies always get deleted ?!?!
Hello Andy, wow and thank you so much for taking the time. (Strangely my answer yesterday was not posted?! )As you write, the result looks pretty good considering the original. I tried myself to sharpen and optimize via RT from the PR3 Output and am very pleased. That should be the way for me to go.
The supposed Redstart (don't know what it is TBH) is of much better quality mainly because of the gathered light. My own editing on that one without using AI tech is quite satisfactory to me. Don't know, if you had a look at the pp3.
I am not sure if the pics are out of focus. I feel that it is more the low light (Shutter speed and available light) in combination with the small sensor (more prone to noise) and blurring by heat haze (which may also throw off the focus). The "Redstart" is F/6.3 1/320s ISO400, while the Peregrine (I think it is a hobby?/falco subbuteo) is F/9 1/1600s ISO 5000 and it is still underexposed cause it was getting dark already. But that was a pretty warm and sunny day so heat haze might very likely have been an issue. The grebe is an old shot from about 10 years ago with a much worse lens/camera combo and I made it from my canoe...
On Linux you can run MacOS or Windows on a KVM based virtual machine. If you got the hardware you can even pass a GPU to the virtual machine and run the VM at near native speeds as if it were running native on the machine. A lot of gamers are run windows on linux with GPU pass through and getting just as good performance, but even if you don't need such power you can still get a basic VM setup to do simple task such as denoise. Another alternative would be to see if one can get the apps running with WINE or Play on Linux but not all apps run well with those emulators.
I've got Topaz denoise (not AI), and Darktable opens the DNG files without a problem. It must be something related to Topaz AI.
That old non-Ai version does not do the same thing as either Denoise Ai or Photo Ai Juan, so you are missing out! The Darktable devs have just had a sample from me to investigate, but have you seen my supplementary video, linked in the pinned comment?
@@AndyAstbury Yeah, that's why I pointed out that has to be something related to the new versions ( I don't have one to check it out myself). I've only tried with the last one I've got.
@@Juan-wu9vb Non-Ai Topaz DeNoise does not do Raw Denoising. If you are on Windows or Intel Mac (not M1/Apple Silicon), getting the DeNoise Ai version 3.7.2 will blow your mind!
@@AndyAstbury I'm using Topaz Denoise AI 3.7
Interesting comparison.
You can run the output DNG from Topaz through the Adobe DNG converter, and it will load in Darktable without any sculls (for some obscure reason) - still the demosaicing is a done deal. The further editing seems to be as smooth as with the DxO DNGs (I only use the latter in Darktable).
You can zoom with the scroll wheel.
Another pleasant tool in Darktable for those who want to use DxO without any sharpening - is the Diffuse or sharpen module. It can even soften sharpening in specific ways and areas. It's not easy to grasp right away, but Boris Hajdukovic has made the following video, which imo is the best to explain this specific tool (observe the ability to blur an edge, while keeping contrast around it):
th-cam.com/video/pAbyORw0mng/w-d-xo.html