I appreciate the dust removal being automized. So far quite useful. One major drawback: Sensor noise is also cleared away leaving noiseless circles and they show. In order to work properly the sensor noise in the area of the removed dust needs to be restored.
I now recall the previous mention of the change to 200 dust spots and that should be enough I think.
Most of my images these days would rarely get to f-stop level where dust would become an issue but my set of trial images, from beta testing days, used a camera that has not has a sensor clean in some years making it an ideal source source of sample images when the lens aperture was stopped down. (Oddly my second body, equally uncleaned and probably with more lens changes since the last clean, shows no signs of any dust spots at all, which is puzzling.)
Anyway, I have counted 47 auto-identified spots on one image. I may have miscounted but at any rate we are nowhere near the 200 limit. Other images taken at the same time show about the same or a little less, so any spots not identified should not be due to a number limitation.
Now, to be fair, many of the spots are edge cases and the images were taken with the specific intention of highlighting sensor dust appearance. So, overall, the Auto detection results are very good.
However, I think they could be better or at least more consistent. So there may still be some worthwhile work to undertake at the detection stage.
Most of this observation is based on the same images that I used for beta testing.
However, with my seemingly "clean" sensor second body, using some very recent images, I sometime see some odd "false positive" spots detected. One image I have looked at today identifies, using the Auto facility, a small spot that does not have any obvious existence at all and a larger area that again is almost imperceptible but looking very carefully at a very underexposed variant of the image (it's a full-on sunset that is not quite burned out) one can see an extremely feint section of backlit cloud that could, perhaps, be interpreted as a dust spot.
In that example the auto-correction makes no difference to the image but I suppose it might have some influence on system performance. It is truly no big deal in this case but could it be more disruptive to process flow in other cases I wonder?
I am wondering whether it might be useful for me to create a Support Case and maybe a short screen grab video to illustrate what I am trying to describe. Would that help?
The change in 16.2.3 only improved the rendering. Currently, there is a limit of 200 dust spots. There are a few reasons for having a limit, both performance and architecture wise in the code. With the introduction of the auto dust removal, we actually doubled the limit from 100 to 200.
We haven't seen many images with more than 200 dust spots but it can of course occur! We will keep an eye on it.
I'm still a little puzzled by the results of the dust point selections, though the blending does look good so far.
Are there any limits to the number of points the Auto assessment will identify?
My test files are still showing the selection of some almost imperceptible dust spots - but not all of them.
Some of the more obvious ones are not being selected even though they appear very near other dust spots that have been selected and successfully treated.
There is probably a good reason for this apparent discrepancy but for the life of me I cannot work out what it might be!
Any suggestions? I am more than willing to test any theories and principles that are in play here.
I now have another good excuse to not invest money I cannot afford into mirrorless technology... ;)
The results with your files look very much the same as my results with my files.
Great concept that, hopefully, be able to rapidly evolve quite quickly. Some use controlled parameters may be useful at some point. To address the trickier choices for "default" settings.
If the new automatic dust removal can clear these it'll be close to perfection. Dust spots over fine structures like twig silhouettes are going to be tough as well. But the mirrorless cameras with the sensor being exposed at every lens change have brought back a problem that was almost not existent anymore.
Thanks for the post! We are actively working on improving this part of the tool. Are you happy with sharing some RAW files we can use to test the tool with internally?
Hi Alexander,
Many thanks for your reply.
I now recall the previous mention of the change to 200 dust spots and that should be enough I think.
Most of my images these days would rarely get to f-stop level where dust would become an issue but my set of trial images, from beta testing days, used a camera that has not has a sensor clean in some years making it an ideal source source of sample images when the lens aperture was stopped down. (Oddly my second body, equally uncleaned and probably with more lens changes since the last clean, shows no signs of any dust spots at all, which is puzzling.)
Anyway, I have counted 47 auto-identified spots on one image. I may have miscounted but at any rate we are nowhere near the 200 limit. Other images taken at the same time show about the same or a little less, so any spots not identified should not be due to a number limitation.
Now, to be fair, many of the spots are edge cases and the images were taken with the specific intention of highlighting sensor dust appearance. So, overall, the Auto detection results are very good.
However, I think they could be better or at least more consistent. So there may still be some worthwhile work to undertake at the detection stage.
Most of this observation is based on the same images that I used for beta testing.
However, with my seemingly "clean" sensor second body, using some very recent images, I sometime see some odd "false positive" spots detected. One image I have looked at today identifies, using the Auto facility, a small spot that does not have any obvious existence at all and a larger area that again is almost imperceptible but looking very carefully at a very underexposed variant of the image (it's a full-on sunset that is not quite burned out) one can see an extremely feint section of backlit cloud that could, perhaps, be interpreted as a dust spot.
In that example the auto-correction makes no difference to the image but I suppose it might have some influence on system performance. It is truly no big deal in this case but could it be more disruptive to process flow in other cases I wonder?
I am wondering whether it might be useful for me to create a Support Case and maybe a short screen grab video to illustrate what I am trying to describe. Would that help?
Hi SFA,
The change in 16.2.3 only improved the rendering. Currently, there is a limit of 200 dust spots. There are a few reasons for having a limit, both performance and architecture wise in the code. With the introduction of the auto dust removal, we actually doubled the limit from 100 to 200.
We haven't seen many images with more than 200 dust spots but it can of course occur! We will keep an eye on it.
Alexander
Alexander,
I'm still a little puzzled by the results of the dust point selections, though the blending does look good so far.
Are there any limits to the number of points the Auto assessment will identify?
My test files are still showing the selection of some almost imperceptible dust spots - but not all of them.
Some of the more obvious ones are not being selected even though they appear very near other dust spots that have been selected and successfully treated.
There is probably a good reason for this apparent discrepancy but for the life of me I cannot work out what it might be!
Any suggestions? I am more than willing to test any theories and principles that are in play here.
Hi all,
The new rendering algorithm for Dust Removal is now released and the feature is no longer in beta. Please try it out, it's available in v16.2.3.
Alexander
Ah okay... much better... to be straight - I still wouldn’t approve it for our library but much much better.
Well done... I don’t get that results on mu version yet...
Eduard
Hi Eduard,
You can right-click the image, click "copy image address" and paste it into a new browser tab. Then you get the image in it's original size.
Alexander
Hi Alexander,
now seeing your result...quite small but is there a way you could share it?
Eduard
Looks really good! Looking forward to try it out myself.
Jochen
Hi Jochen,
Just to give you an update on the progress, here's an example of the current Dust Removal with the visible traces compared to what we are working on:
Alexander
Thank you Jochen!
I now have another good excuse to not invest money I cannot afford into mirrorless technology... ;)
The results with your files look very much the same as my results with my files.
Great concept that, hopefully, be able to rapidly evolve quite quickly. Some use controlled parameters may be useful at some point. To address the trickier choices for "default" settings.
Here are two files for you to test. Enjoy!
https://magentacloud.de/s/MT5TXSDAZ8oPq28
If the new automatic dust removal can clear these it'll be close to perfection. Dust spots over fine structures like twig silhouettes are going to be tough as well. But the mirrorless cameras with the sensor being exposed at every lens change have brought back a problem that was almost not existent anymore.
Jochen
Hi Jochen,
Thanks for the post! We are actively working on improving this part of the tool. Are you happy with sharing some RAW files we can use to test the tool with internally?
Alexander