Thought I would try to use the AI mask to alter a near blown out sky but as before, it bled into the background and didn’t work at all.
Just out of curiosity, I thought I would export it and see what the jpg looked like. It failed to load. I had recently used DxO 9 to deal with 40 or so photos (no AI) and they worked fine. But as soon as an AI mask was used - no export.
I brought up the Radeon Software monitor and tried the export again. It got 3/4 of the way through during which time the GPU spiked a couple of times, but the VRAM, RAM and CPU never got anywhere near max. This is the norm for exports without AI masking being used.
The only good news is that the AI masking doesn’t really work very well, if it did, I’d be peeved.
If you mean the preset AI sky mask, then that (along with the other AI presets for background, subject, etc.) is a known issue. The AI presets appear to be barely at beta stage.
If though you mean the manual selection of the sky by AI failed then that’s a surprise as many people here, myself included, don’t have problems with manually selected AI masks.
Thanks for that Joanna. I think it was more to do with the fact that DxO and users should be aware of issues before they dive in feet first.
I don’t use the local adjustments much. There is a balance between the time taking photos and the time processing them. I have al;ways been very conscious that every minute I am faffing around on a PC is another minute spent not taking a photo. Better to have a not quite perfect photo than not have it at all.
My curiosity is what is making the software fail. As it appears the vital components are not maxed out in any way. So if it’s not the hardware……………………….
Manually selecting isn’t the problem, although any resemblance of accuracy certainly is. After several attempts it was obvious that the alteration to the sky bled into other areas or at least appeared to.
The selection isn’t the problem it is the export. As I don’t see the AI masking very well in the two cases I have tried, it’s not a big deal for me. But if the hardware is not maxed out then what crashes the export?
Export works fine when AI local adjustments aren’t being used.
What is the GPU VRAM amount? 6GB? 8GB? NR is DP3 or DP XD2s? Any AI pre-defined (keyword like: sky) mask used? Deeprimerendering is On? Loupe used before Export?
GPU not use the ‘normal ram’ (shared) at general - especially for AI model (if you select the GPU in the preferences)(. You already experience that → as if its can use as ‘shared’, than no issue raised expected.
Seems the pre-defined AI mask (‘Keyword’, like: Sky) not friend with 6GB GPU VRAM + Exporting. Actually i think in his forum near all ‘AI mask crash/mask Internal error/Export error’ related with that Pre-defined (‘Sky’) vs GPU VRAM amount. Practically all AI / NR / Export use some GPU VRAM amount. For example (to get the concept): ‘Manual AI mask’ AI model use less GPU VRAM than ‘pre-defined Sky AI mask’. DP3 NR use less the DP XD2s, and so on. I have no idea about X-trans usage, but i guess its same or more than DP DX2s. Loupe / DeeprimeRendering also also use some GPU VRAM amount, etc. And Export use ‘again’ - as you can edit while export running → export need to ‘run again’ the ‘pre-defined Sky AI model’ (at least some similar). And you simply ‘runs out’ for GPU VRAM (during export).
If you use manual AI mask (selection, area) and not use ‘pre-defined AI’ → you may just fine in Export. May use DP3 (if possible, i’m not familiar with Xtrans) may more in the safe side.
May worth to try this export small ‘trick’!:
It’s may also help.
Deeprimerending?
Its like ‘loupe’ in the full editing screen. A new option to view the effects of global and local DeepPRIME denoising in the editing window, full screen and at all zoom levels, while editing.
What is means ? The ‘really usable’. I just wonder on that.
At least i not experience issue with that, that’s why i start to wonder, what its means?
May describe it with screenshot?
On a lot of machines the loupe runs at 100% ok. There is a delay but it’s ok. It is basically the same as the magnifier in PL7, just slower. Anything over 100% just cycles and never renders. Tbh, this doesn’t effect me much, others more so.
I not read too much about that in the forum (loupe issue), but i get it. And if i ‘run out’ of GPU VRAM, i experience some similar with 4GB GPU VRAM (loupe with DP3 is okay, but with DP XD2s nop). I guess the order of steps may can be important - at general: loupe before export or after export, and DP3 vs DP XD2s.
I suggest to try the followings to break down the thing:
Basic test:
Start PL.
Get one (1) photo. Add basic adjustments, no mask. NR: Standard.
Quit from PL.
Start PL - its open the last photo (the edited) - not click to any other photo!
Try loupe. Zoom it over 100%
If its works (with Standard NR) than change photo NR to DP3.
Quit from PL.
Do the same sequence in the same photo (step 3- 7 - now with NR: DP3)
See its works or not.
And another round try with DP XD2s.
I expect at least its works with DP3.
If all works, than repeat this steps (sequence) with the following add-on:
‘Advanced test’
Test sequence 1: Basic cycle but Do loupe stuff AFTER Export this photo. - I expect at least its works with DP3.
Test sequence 2: Add manual AI mask (1 is enough) with some adjustment to the same photo - in the next this one is your test photo now. Dont forget to quit from PL. And do the basic test cycle. I expect at least its works with DP3.
Test sequence 3: In this manual AI mask photo and do loupe AFTER Export
May you can do in ‘reverse’ order: Loupe and AFTER export.
Do not click to any other photo! Not forget to quit from PL!
May its looks a lot, but in the reality if just a few simple things. And i think it’s helps to find out the way how you can use it (loupe) safe and sound.
Any question, let me know.
Applying the Mask is not the issue, neither are doing any adjustments that were available in previous issues.
It would be interesting to see how many people can get the loupe to work at 400%. Certain 200% is too much for my machine and this fits within the PC parameter set by DxO. These requirements are not sufficient for working with the AI mask, and I am sure DxO know and knew this.
The interesting thing is that apart from 2 spikes in the CPU for fractions of seconds, the VRAM never maxes out and yet anything involving the AI mask crashes. The VRAM also has 48gb of RAM as back up which the card can utilise if required.
This is not a VRAM issue and the PC configuration is better than the minimum requirements.
It’s almost certainly a software issue (DxO or nVidia or both).
On my aged system, competing software that include AI masking and detection tools (Lightroom, Photoshop, Topaz etc.) do not crash at all. Some might be slow with initial detection, but they do not crash.
PhotoLab 9 - when it first came out - crashed every damn time. Now it only crashes some of the time. My PC specs didn’t change.
But not everything can utilize that ‘backup’. Lot of things cant.
Anyhow, if you share one of your RAW file with the related .dop file, i take a look what happen in my PC with ancient 4GB VRAM GPU. May we find out something. I’m happy to check.
With 5 AI masks being used, and DeepPRIME XD3 enabled, I can use the loupe at 400%, 800%, 1600% no problem. It takes maybe half a second to load any new section I move the loupe to. (I have high quality previews enabled, and DeepPRIME rendering enabled for previews as well).
Interestingly once “full preview” for an image is completed on my machine, I can move the loupe around the image almost instantly - dare I say less than half a second?
BUT…
I just tried making a basic edit (increasing midtones in the main image panel, not a mask) to that same image and timed the “full preview” creation time.
19 seconds.
That is ridiculous. It only takes marginally longer (27 seconds in this case) to export the damn image entirely.
For reference this image has 2x AI mask, 1x graduated filter, 1x auto-brush.
Turning off masking reduces the wait time doing the same test to around 1 second. The “full preview” generation message doesn’t even have time to display.
Turning masking back on again but deleting the AI masks, then making the same test edits to the general mid-tones, results again in a barely perceivable delay and again the message doesn’t have time to display.
I have experienced slowdown with masking in general enabled (especially many masks) but AI masking is definitely in a bad place right now.
Just a note: If you enabled the ‘DeeprimeRendering’ → than the ‘NR section’ selected NR applied to the screen. If Standard selected → Than ‘Standard’ used for screen NR. If DP3 than DP3 and so on.
And ‘what type’ of AI mask? Selection? Area? Pre-defined? Note: ‘auto-brush’ also AI mask as i see. I guess, its Pre-defined AI mask? As in the later its may not surprising.
I think all of this type of things may good ide to share the raw and the related .dop. Give a chance for others to check. May we find out something.
With (now) 6 AI masks – I used 2 predefined and 3 “add selection” and 1 “add area,” and DP3 turned on, I can make most basic edits which render in approximately a half second or less, with the maximum seeming to be one second. Doesn’t matter if it’s midtones, clearview plus, exposure, HSL or curves.
It also didn’t seem to change the timing much whether or not I had DeepPRIME rendering or high quality previews checked “on.”
Export took 31 seconds. (THAT is ridiculous to me.)
Loupe is still responsive down to half second or less, no matter where I put it, and no matter whether I have DP rendering / HQ previews check on or off.
