DXO PL6 vs Topaz AI

Horacecoker said:

tbcass said:

Horacecoker said:

Digital Nigel said:

Horacecoker said:

Digital Nigel said:

Horacecoker said:

tbcass said:

I posted this on the photo retouching thread but because nobody seemed interested I will post it here.

The following are processed RAW from the latest versions of both. They are ISO 5000 from a Sony RX10iv which push the limits of the 1" sensor. Comments welcome.


DXO


Topaz

Click to expand...

I prefer the Topaz image. The artifacts produced in the DxO image are unacceptable to me.

Here's a 100% crop example.

Click to expand...

Yes, that suggests use of DeepPRIME XD without turning down the Noise model.

Click to expand...

But there's another problem. DeepPrime XD looks like it has tried to sharpen these background faces that are obviously out of focus due to the photo being taken at 600mm equivalent zoom. Whereas Topaz AI has rendered them correctly.

Click to expand...

Yes, agreed. I wonder if different settings in PL6 might have produced a better result (eg, using DeepPRIME, not XD)?

Click to expand...

Quite possibly. Perhaps Tom could make the raw file available?

Click to expand...

I have no way of making the RAW file available but when I have a chance I will post a few samples with different settings.

Click to expand...

That's a pity. Would have liked to give it the once over with Adobe Denoise in Lightroom.

Click to expand...

Yes, that would be useful.

Horacecoker said:

A basic Dropbox account costs nothing and you get 2gb of cloud storage.

Click to expand...

I think Tom is referring to the people in the image, not the mechanics of sharing.

Actually I don't understand the comparison: PL6 is a full image editor including the DeepPrime (XD) NR options and sharpening models , while Topaz is no image editor (NR and sharpening only (and upscaling)).

For a fair comparison I would

• process the image first in PL6 doing all the tonal/color adjustments to taste and apply DeepPrime (XD) to taste -> export as jpeg to disk
• then turn off NR in PL6 -> export to Application, i.e. Topaz Photo AI; turn off additional sharpening in Photo AI, run photo AI -> return tiff inside PL6 again, the export the image as jpeg to disk
• Compare the to versions of the image.

Btw, since using PL6 for my RX10iv I actually use Topaz only for sharpening selective parts of an image, if needed. You can do that either within PL6 or on the resulting jpeg from PL6.

Horacecoker said:

sybersitizen said:

... An interesting observation: The Photo AI version shows a couple of stuck white pixels. PhotoLab's dead pixel correction found them and fixed them.

Click to expand...

Out of curiosity I searched and found the two 'stuck white pixels' (it's raining outside) and zoomed them up in Photoshop. In both cases, the defect spreads out to surrounding pixels. Would a single dead pixel produce that effect?

Click to expand...

I don't know. This is the first time I've ever noticed any in my normal work.

Horacecoker said:

In the first crop, the brightest pixel isn't white, it's a very pale blue-green.

Click to expand...

Well, white was just a convenient approximation. There's nothing actually white in either.

Horacecoker said:

Click to expand...

tbcass said:

sybersitizen said:

I've been using both DxO PhotoLab and Topaz Photo AI for longer than you have, and that doesn't look like a characteristic Topaz Photo AI result. It normally removes nearly all traces of noise. Often I choose to override the Autopilot settings to allow a little noise back in ...

Click to expand...

The fact that you have been using Topaz longer than I have doesn't matter because the latest versions of Topaz make earlier versions obsolete with all the recent upgrades.

Click to expand...

I'm not talking about different versions. I'm talking about potential differences in the way we use these tools being the relevant factor.

tbcass said:

I have to upgrade Topaz about every 2 or 3 weeks.

Click to expand...

I go even longer between updates. It's annoying to keep updating frequently.

tbcass said:

Previously the difference between Topaz and Photolab was very obvious but with the latest upgrades I installed a few days ago, they are closer.

Click to expand...

EXIF says the image you posted was done using v1.3.11. My version was done using v1.3.8. My 'older' version produced less difference than yours, not more, so that's not a factor here.

tbcass said:

DXO however is my go-to RAW editor because it's full-fledged whereas Topaz has limited controls and relies more on AI for everything.

Click to expand...

Same as what I said. But I suspect that I've dug into Photo AI further than you have, which was the point of my comment.

tbcass said:

I never noticed the dead pixels because I tend not to pixel peep that closely. could it possibly be an artifact from the processing not present in the DXO version?

Click to expand...

Both of the errant pixels are in the ARW file, but they hardly stand out at all. Now that the month has rolled over, I expect that the camera has performed its monthly pixel mapping. I'll have to do another test and see if those two still turn up.

sybersitizen said:

Horacecoker said:

sybersitizen said:

... An interesting observation: The Photo AI version shows a couple of stuck white pixels. PhotoLab's dead pixel correction found them and fixed them.

Click to expand...

Out of curiosity I searched and found the two 'stuck white pixels' (it's raining outside) and zoomed them up in Photoshop. In both cases, the defect spreads out to surrounding pixels. Would a single dead pixel produce that effect?

Click to expand...

I don't know. This is the first time I've ever noticed any in my normal work.

sybersitizen said:

In the first crop, the brightest pixel isn't white, it's a very pale blue-green.

Click to expand...

Well, white was just a convenient approximation. There's nothing actually white in either.

sybersitizen said:

Click to expand...

Click to expand...

I assume that, as part of the demosaicing process, a single dead pixel in the sensor affects a cluster of image pixels. It's not a 1:1 relationship.

Digital Nigel said:

I assume that, as part of the demosaicing process, a single dead pixel in the sensor affects a cluster of image pixels. It's not a 1:1 relationship.

Click to expand...

That's probably right, and it could also explain why the central pixels are not actually white after demosaicing.

sybersitizen said:

Digital Nigel said:

I assume that, as part of the demosaicing process, a single dead pixel in the sensor affects a cluster of image pixels. It's not a 1:1 relationship.

Click to expand...

That's probably right, and it could also explain why the central pixels are not actually white after demosaicing.

Click to expand...

Yes, by definition, a single dead pixel can't be white.

Talk of dead pixels made me look at DxO, in my case Photolab 5.12.0 (despite being "old" it does get regular updates for new cameras and random fixes).

I found the dead pixel control....



I needed to click "Advanced Settings" to see the extra sliders, shown at default for one of my early RX100M6 shots.

Never noticed a dead pixel anyway with my RX100M6 but then, never really looked. My various Olympus bodies have a user menu to do dead pixel mapping and I do that every couple of years whether needed or not.

The dead pixel slider in DxO possibly plays havoc with astro photos?

Guy Parsons said:

Talk of dead pixels made me look at DxO, in my case Photolab 5.12.0 (despite being "old" it does get regular updates for new cameras and random fixes).

I found the dead pixel control....



I needed to click "Advanced Settings" to see the extra sliders, shown at default for one of my early RX100M6 shots.

Never noticed a dead pixel anyway with my RX100M6 but then, never really looked.

Click to expand...

Presumably DxO detects and hides them?

Guy Parsons said:

My various Olympus bodies have a user menu to do dead pixel mapping and I do that every couple of years whether needed or not.

The dead pixel slider in DxO possibly plays havoc with astro photos?

Click to expand...

I'd have thought not. A single red/green/blue dead pixel won't look anything like a distant white dot.

Digital Nigel said:

Guy Parsons said:

Talk of dead pixels made me look at DxO, in my case Photolab 5.12.0 (despite being "old" it does get regular updates for new cameras and random fixes).

I found the dead pixel control....



I needed to click "Advanced Settings" to see the extra sliders, shown at default for one of my early RX100M6 shots.

Never noticed a dead pixel anyway with my RX100M6 but then, never really looked.

Click to expand...

Presumably DxO detects and hides them?

Digital Nigel said:

My various Olympus bodies have a user menu to do dead pixel mapping and I do that every couple of years whether needed or not.

The dead pixel slider in DxO possibly plays havoc with astro photos?

Click to expand...

I'd have thought not. A single red/green/blue dead pixel won't look anything like a distant white dot.

Click to expand...

I'm not an astro shooter so was just guessing, here's a post from someone about an astro shot with DxO https://www.dpreview.com/forums/thread/4233948

sybersitizen said:

tbcass said:

I never noticed the dead pixels because I tend not to pixel peep that closely. could it possibly be an artifact from the processing not present in the DXO version?

Click to expand...

Both of the errant pixels are in the ARW file, but they hardly stand out at all. Now that the month has rolled over, I expect that the camera has performed its monthly pixel mapping. I'll have to do another test and see if those two still turn up.

Click to expand...

Well, I did my test.

This is my previous example done with Photo AI:


Topaz Photo AI with Autopilot settings (Remove Noise: Strong, Strength: 8, Detail: 82)

This is the new example with the same capture settings and the same processing settings:


Same settings as above

The original hot pixels have disappeared, which I suppose might be the result of today's new pixel mapping. But now there are different ones. Look at the black guitar at the left, for example. I don't have an explanation for the new ones suddenly appearing.

Digital Nigel said:

I suspect DeepPRIME might have been the better choice with this example. And you should always turn down the Noise model with XD (I suggest -40).

Click to expand...

I will take that under advisement. I tried Deep Prime first but felt XD gave better results in this case.