Lens Sharpness Optimisation changes in masks causing severe in-program slowdown

DxO PhotoLab
1

Credit where it’s due; PhotoLab 7 performance has improved since launch - yes even for me - and I’m glad to see that’s the case. It’s almost in a good place, even on my old system.

There is one little gremlin however, and I think I’ve tracked it down.

I speak of Lens Sharpness Optimisation changes, specifically made using masks.

If I set LSO as a global value, there are no issues. I can make edits globally or using masks without significant performance loss (too many masks and things do slow down, but that’s expected).

However, my preferred editing process is to apply a low LSO value globally and apply a higher value to specific subjects via masks.

If I do that, PL slows to an absolute crawl and takes half a minute - maybe more - just to generate its “full preview”:

FPIP
FPIP285×196 20.5 KB

This performance loss seems especially bad if the LSO change is applied to an AI-generated mask, but it still takes a while longer than other edits to apply to non-AI masks (control points, auto-brush etc.).

I don’t see the same slowdown with changes to exposure, tone, ClearView, contrast, WB, HSL, denoising or blur. It’s strictly local masked LSO changes that causes this.

I’m not sure there’s a “cure” for this at the user end, except to implement local LSO changes last before exporting… which is what I do currently. It’d be nice if it didn’t have such an impact, of course.

1 Like
2

I’m guessing you meant to type 9, not 7 (?)

I reckon DxO Support would value having this feedback (they won’t necessarily see it here) … Esp. as you report performance is otherwise good.

1 Like
3

Aha ! … I’ve just encountered this issue too - - with LSO correction applied locally.

I was surprised to see the error message below, and then I recalled your post.

  • image

Definitely worth reporting to DxO Support … Are you doing that ?

  • If so, I suggest providing a link to this thread.
4

You’re right - I did mean version 9 :sweat_smile:

I’ve not reported it yet but will do so now, citing this thread. Good idea.

2 Likes
5

Not on my equally old PC, which I equipped with a NVidia RTX 4070 12 GB to extends it’s life.

And there are many ways to “control” sharpness, e.g. locally

vs

Note – this example is for demonstration purposes only.
(May be helpful for birdwatchers who need to crop heavily.)

2 Likes
6

No such specific problem here with Win11/RTX4070/i7-14700KF/32GB RAM/45mpx Z8.
I often use LSO negative local settings with AI masks together with strongly negative Microcontrast to tame the background, so I would have noticed that.

If I get a slowdown, it’s because of prolonged AI masks usage, interleaved with exporting. Correlated with large ‘Shared GPU Memory’ usage, cured by PL restart. This happened MUCH more often with pre PL9.6 versions but it can still happen with 9.7.

7

Same on my side - no problem using PL 9.7.0 with Win11/RTX3070/AMD9950x3d/64GB RAM/45mpx Z7II.
LSO for the Person-AI-masks was set to 200, rendering of image needed about 2s. Export (to tif) also went flawless and took approx. 10sec (still too much time but probably due to weak 3070 GPU).

Stephan

image
image1920×421 204 KB

8

Minor update; I’m still going through the support process, provided RAW and .DOP file along with detailed instructions.

I was also asked to run a diagnostic tool - having done so I found this doesn’t just look at hardware and DxO software but also scanned (and packaged up for DxO) some residual folders I had for Adobe products too.

Not really comfortable with DxO software poking around at anything besides DxO so I’m not sending over that report.

Interesting; thanks for that - our systems are certainly similar(ish) in terms of year and horsepower besides that 4070.

I just did a quick and brief search for 4070Ti prices second hand and on CEX they’re still showing as ~£550 or more which is not inconsiderable.

I’ve found increasingly LSO a little to be a pleasing way of controlling sharpness which is why it’s frustrating that it tanks performance on my system! I think your own example is showing that too, as your “very soft” example shows LSO as “off” and DeepPRIME XD3’s forced details as actually negative 100(!)

Compare that with your extremely sharp example and you’ve got LSO overclocked to 200 and XD3’s forced details at +100 too.

Interestingly I’ve not tried negative values e.g. starting with a very sharp image and selectively making parts of it less sharp. I wonder if that would impact performance in the same way as additive sharpness does.

9

As said, the example is for demonstration purposes only.

To make things perfectly clear, I have used maximum values, which you would not do if you were editing carefully.

10

I also experience this (slowdown on mask LSO changed)! Its definitely slow down things, in Editing for sure (i not measure Export) even if its on “standard” (non AI) masks. I guess in powerful computers is not “feel” to slow things down, but less powerful its slow. And seems very low CPU/GPU usage when “calculate” LSO in this case(s). I also wonder to write a post about that!

p.s.: " auto-brush" → its AI.

1 Like
11

This makes me somewhat gratified I’m not bizarrely alone or imagining it!

As you say, it’s not during export (I’ve not noticed any great change there, although exports are faster the fewer options one has enabled and so I imagine adding more masks lengthens the overall export).

The point/problem is the performance drop while editing, due to enabling LSO in a local edit/mask.

And as you say, presumably faster systems are just that - faster - so there’s little/less/no issue enabling local LSO changes there. Unfortunately “just get a faster system” isn’t necessarily cheap.

1 Like