Prime NR vs. than Fast NR with pushed high ISO

here the Fast NR image:

Hello @obetz,

can you send us your RAW image, if that’s not a problem.
I’d like to see if we can improve denoising.

regards,

where shall I send it, as an attachment in this thread or somwhere else?

Hello Marie,

you can retrieve two sample images from box.com:

https://app.box.com/s/lwycl1z6jqyhckynly5hsg3g2dcd8siq the source of the jpg file mentioned above

https://app.box.com/s/cmmizzq6gmjsd7eftfeqbm6dgv8itkus another sample, look at the tomato

Please let me know if you got the images.

Oliver

Thanks, I got images. We’ll analyze that.

Hi Oliver,

As a Sony owner, you’ll probably be interested in this: BUG: PL Smart Lighting has lost algorithm to process (purposefully) under-exposed images from ISO-Invariant sensors

Hi John,

it’s about noise reduction, not about selecting the right exposure. Therefore it’s not related to the topic you mention.

Nevertheless I would appreciate better control over handling “purposefully under-exposed images”, e.g. a separate control where I can set a percentage of camera setting to be pushed by DxO before applying smart lighting.

YES !!

How about raising a request/suggestion in the Which feature do you need? category for PhotoLabI would certainly vote for it.

Regards, John M

I have a very different experience with DXO - possibly it is your sensor?

Moonlight - significant under exposure: D300; 15 seconds; f/5.6; ISO 1600 - “Prime” de-noise.

For my D300 - 1600 is high ISO (the camera was developed in 2006) - 3200 also works - 6400 is fantasy land.


Original exposure:




Deliberately uber-pushed to “daylight” as a demo:

Hi Jim,

Yes,that’s a good example of underexposing at time of capture - and then “brightening” the image during processing … as a very useful approach for high-contrast scenes - and especially applicable to cameras with ISO-invariant sensors.

However, the problem is that the Smart Lighting algorithm in OpticsPro (via the OpticsPro9 mode of the Uniform method) produced a FAR, FAR better result, and a lot more easily than we are now able to get out of PhotoLab for the same underexposed images (since this algorithm was dropped from PL).

For more background on this, see here; Bemoaning loss of DxO Smart Lighting - Mode = OpticsPro9

And, to add your support to resurrection of this lost capability, see here: BUG: PL Smart Lighting has lost algorithm to process (purposefully) under-exposed images from ISO-Invariant sensors

John M

Hi Jim and John,

may I remind you that the smart lighting issue is completely off-topic here. New postings should only be added to this thread if they are about problems of Prime NR with very very dark scenes.

@obetz
Sorry for mis-reading the intent of this thread.

EDIT: My off-topic post that @obetz referenced has now been deleted.

Attached is the same image posted previously (D300), this time with only distortion correction & Prime noise reduction in PhotoLab v1.2. As per my first post - I think this is a sensor issue… i.e. the Sony RX100 IV has a uniquely stacked sensor (Exmor RS stacked CMOS) that is not being interpolated by PhotoLab correctly.

As evidenced by the following images the issue You described is not universal.

Exposure: as shot; 100% crop; click to enlarge.




Exposure: 2x push (as per your original post); 100% crop; click to enlarge.

@obetz I had a look to images you sent us.
First I want to explain that PRIME NR is better than Fast NR on preserving details and colors.
On your images we can see that residual chromatic noise is specific to corners of images, dark areas like corners but on other areas don’t have that problem. We don’t have that behavior on our tests images.
So to me it’s not a problem of calibration but you may have an issue on your camera.
Nonetheless I will discuss that with Image team to see if we have some possibility to correct it.

Hi there,

I have the same problem as obetz. I am using a Lumix G9. The following picture was taken at 200 ISO with a 50 s exposure time, quite long, I know, and I admit it is not a pushed high ISO as in the example above but it shows the same behaviour. However, the fast noise reduction gives an overall good image, whereas the Prime Noise Reduction is superior in the sky and the structure of the building, but the foreground looks worse with far more noise than the fast method and almost no structure visible. Attached the photo and an excerpt with the fast NR.


… and here is the photo with the Prime Noise Reduction.


It’s easy to tell so, but you don’t know whether it’s a Prime NR issue. May I suggest you grab another RX100 IV and take dark images with ISO6400 pushed by two steps to see whether you are right?

Hello @obetz,

I may have answer you too quickly (we just started to look at your issue). We are working on RX100 IV and we will be able to reduce the problem :slight_smile:
But there will still be some difference in corners even if result should be much better.

any news on this issue?

Hello @obetz,

sorry i forgot to update that topic.

We have increase denoising of chromatic noise with PL2 for high ISO on RX100 IV, that helps but doesn’t solve fully the issue as there is something odd (noise is too different between areas of same luminance).
You can try trial version of PL2

Best regards,

Postponing this to the next paid release is not customer-friendly.