New Version - Old Problems

I reported significant performance issues with the previous version.

Over time, the thread morphed into an AI-mask thread. I’m now sharing my experiences with the new PL9.2 in a new thread.

After the support team’s incredibly attentive service, I would have expected some kind of response from the development team.

Every problem was acknowledged, my patience and valuable cooperation were praised, and I was constantly asked for new data.

Looking at the new version now, I’m certain I’ve been strung along by AI. Call me ‘naive,’ but I actually know of companies where a bug report actually prompts a response and fix.

To edit images in a specific folder, that folder needs to be selected somehow. Windows provides standard dialogs for this. Opening a directory and displaying the names of the folders and files never takes more than a second. This applies regardless of whether it’s images, music, or text, even if the data is on a slow hard drive.

I respect the developers’ desire for a more visually appealing interface than the Explorer. However, if the custom folder selection on a hard drive takes a minute to display a folder with several thousand subfolders, then the programmers’ skills clearly lie elsewhere.

The idea may have been to examine all other subfolders before the user opens them. Even if this worked without any issues, I don’t see any significant advantage. If this behavior slows me down, then it’s unusable.

Eventually, the directory list appears, and I can navigate to the desired subdirectory. There are also many files there. I can scroll through the thumbnails, even if they aren’t fully loaded yet. If I move the scrollbar, the corresponding thumbnails are immediately loaded. After that, the process continues in the background for all other thumbnails in the folder. That’s how it should be.

Now I’ve edited my image and switch to a folder with 200 images on the SSD. At first, it looks perfectly normal. PL9 now recognizes that a new module is required for a new camera/lens combination. The download only takes seconds. Afterward, however, PL9 is blocked for several minutes. The cancel button is also ineffective.

During this waiting period, I start ProcMon and record which file accesses PL9 makes during this time. PL9 is now reading all the images from the folder previously opened on the slow hard drive. I have no idea why—I’m already working in a different folder on the SSD. Perhaps to finish building the thumbnails? This isn’t desirable in my opinion; I might not even open this folder again in PL9. Most importantly, this operation shouldn’t prevent me from continuing to work with PL9.

I have described this behavior to support several times. Their follow-up questions then focused on network speed, hard drive type, and GPU. All things that are irrelevant here. Clearly, a local SSD is faster than a hard drive in a NAS over a slow network. All of this is more than fast enough to browse a folder in Explorer and load an image into a photo editor. However, slower hardware is better suited to revealing inefficient processes within a program.

It’s a shame. As good as the image editing results are, the overall package is unusable.

2 Likes

Performance is at an all time flow for me with the latest version (9.2.1).

I’ve downgraded to the previous version (which forced deleting PhotoLab and then re-downloading a whole library of AI masks, irritatingly) but that at least seems to load folders quickly (for PhotoLab) and export in something vaguely acceptable (30-35 seconds per image, vs the 2.5+ minutes I’m seeing with 9.2.1+).

Then there’s the question mark over why PhotoLab needs to generate a brand new preview when navigating of image A when navigating off image A to B… and then back to A… without any other changes being made. I’m assuming it doesn’t cache anything from one moment to the next.

Folders load fairly quickly for me but - yes - thumbnails seem to need to be generated every time, and the program is sluggish and unresponsive for editing until this process is complete too.

May i write the ‘original’ thread, PL9 only check the ‘file list’ (filenames, creation dates) in ‘all’ sub-folders. Its quite fast (on my measurements) (i not try with NAS, only local SSD).

During this waiting period, I start ProcMon and record which file accesses PL9 makes during this time. PL9 is now reading all the images from the folder previously opened on the slow hard drive. I have no idea why—I’m already working in a different folder on the SSD. Perhaps to finish building the thumbnails? This isn’t desirable in my opinion; I might not even open this folder again in PL9. Most importantly, this operation shouldn’t prevent me from continuing to work with PL9.

I think its not generate the thumbnails for the folder. Its generate only for photos in the ‘filmstrip’. Anyhow its still fast as i see (on local SSD, but i think its can be relatively fast on ‘slow hdd’. Update: may its changed in latest PL version vs. when i test in the past!!!

File ‘readup’ (for photo metadata, camera body+lens combination, etc) not ‘background’ process. I guess its need to wait ‘until its finish’.

‘I might not even open this folder again in PL9’ → but you may open again. And you may also can use ‘search for images’.

During this waiting period, I start ProcMon and record which file accesses PL9 makes during this time. - if you send the ProcMon dump file via some Gdrive, Onedrive, i check it (when i has a little time)

I think its some way logical, and save a time! When you click to a photo (‘A’), its ‘render’ the ‘high quality preview’. That’s the point (in time) when photo is ‘fully loaded’, and the ‘best time’ to generate the thumbnail when you click to another photo (‘B’) - as photo ‘A’ is fully rendered. And seems its the ‘fastest’ way to generate the latest thumbnail. Its save time to see the latest photo version ‘thumbnail’. Other apps work more-or-less in the same way (click from A > B)

‘without any other changes being made’ → Yep, but i think thumbnail generating is quite fast, may doesn’t really matter.

Sorry if I wasn’t clear, I mean the full image gets generated when clicking from A to B to A again.

So starting on image A, that’s:

  • Generate a full preview of image A (make no other changes).
  • Click on to image B. Generate a full preview of image B (make no other changes).
  • Click back to image A. Generate a full preview of image A again.

That’s 3 full previews being generated, even with no changes made to either of the two images.

If changes already exist (masks etc.) this can take time, but even if they’re both “new” there’s a stutter.

If I view raw image files (with no changes) in Photo Mechanic or IrfanView, I can skip back and forth far more rapidly.

1 Like

That’s an unfair comparison because IrfanView is simply showing you the embedded JPEG, it’s not actually doing anything with the RAW data.

1 Like

Because they not show the editing.
Some (most) of the viewers use embedded jpg (Lr also has an ability to that as far as i remember in Library or what mode) - and its very-very fast. Some of the viewers also create 'thumbnails cache file (XnView can as i remember), or cache to file/db 1:1 preview in some way. In the XnView if you not use ‘browsing’ but open photo, its say ‘Picture converted ot 8bit RGB’ and may does some basic RAW rendering (may)

Based on my previous observations:

  • PL not save (‘cache’) full 1:1 for load.
  • PL create like 4 full ‘image’ during ‘full preview in progress’ rendering: without anything (no geometry), with geometry, with standard editing, with local adjustments. And in ‘compare’ its does twice.
  • PL create to the ‘preview cache’ folder only small jpg, like 1100x800 or something (i write a long comment on that once). Its a ‘dummy’ ‘cached image’. Only used (at least in my observations) to display something until ‘full preview in progress’ done - practically use only for ‘fill the void’ until ‘full preview in progress’ done

PL may (at least i not observe that) not use ‘memory caching’ for A → B ->A → B like operations (like no memory caching for A and B). May its has something, but i not observe. That the reason why its ‘full preview in progress’ happen each time. Also to see photo in all ‘edited details’. Not just some ‘jpg’ quality zoom to 1:1.

Anyhow, after A-> B-> C-> D … for lot of photos (may even in a handful amount) also raise memory issues if implemented. And anyhow if its use something, if you do any editing, its need to run again ‘full preview in progress’.

So, at overall its cant be applied in an easy way any ‘trick’, if any good way. Actually i think is pretty forward way the actual ‘full preview in progress’. At least in ‘Correction’ mode.

Assuming I’m starting with a blank slate - no edits to my images - I can’t see why PL can’t do similar. It certainly shouldn’t need to re-generate the preview of an unchanged image from one moment to the next.

(Capture One is also crazy fast at this, so I simply don’t believe it’s not possible to do better).

Appreciated, but I’d quite like the ability not to do that, or to control its frequency or fidelity. I don’t need absolute quality in editing, only in export. In editing I need speed and responsiveness.

1 Like

Neither can I but have you not been around this forum long enough to understand that DxO do things their way in their own time? So much so, they are masters in the art of annoying their customers.

1 Like

No arguments there from me, although it’s nothing personal - I hate that you’re right!

2 Likes

Unfortunately it’s unusable at high ISO, so I have to feed it with RGB TIFFs from PL :frowning:

1 Like

Capture One use some medium quality, and not 1:1 as ‘preview’ but some medium resolution (as i remember, you may correct me).

If the ‘blank’ state means 'not apply any editing, any preset, etc. - PL i think pretty fast. Geometry correction may add something.

I think things i wrote upper is ‘conjuction’, like its need 4 image (/wo Geo, /w Geo, /w editing, /w Local), etc. And not just because its ‘not full quality 1:1’.

But anyhow, i get you.
Regarding ‘fidelity’ and other ‘reduced quality’ based is hard to say what can be ‘good enough’, how to implement (in reliable way) and so on.

Without AI masks in my opinion PL is pretty fast on this ‘full preview in progress’. AI mask, that!s where things start.

In editing I need speed and responsiveness.
And that’s what you get after the ‘full preview in progress’! You may wait in the ‘beginning’, but after editing is speedy! Of course its a standpoint of view.

Also, may also depend on users workflow. I have a pretty slow computer. For culling (pre-select) for 10-20 photo i use PL. For 20-50 i use XnView. If more photos or ‘more precise’ culling need, i use ExposureX. And only culled/pre-selected photos goes to PL. Usually not really more than a hundred to one folder. Also to different folders like ‘Concert1 - Band1’, ‘Concert1 - Band2’. Or even more folders like \Concert1\band1\backstage , \concert, \lead singer and so on.

p.s: all is pros and cons.

1 Like

Another thing I can’t argue with (though I haven’t really thrown much high ISO stuff at it, I do appreciate that’s somewhere PhotoLab outshines anyone).

I did find Capture One responsive in everything though, and a bit more intuitive too. If DxO could make PL 10.0 a performance and experience polish version, I’d genuinely applaud them for that. PL9 is generally feature rich as it is).

Off-topic:

But don’t get fooled, these are not full quality previews in C1. It doesn’t matter, if you have good input, though. For my PC, if I switch off DeepPRIME previews in PL, I get similar “speed feel” for C1 16.7.1 and PL 9.2.1 (and C1 is definitely not crazy fast, you still have to wait just a little for some changes to fully appear). But yes, C1 has done a lot to make 16.6 faster than its previous versions. Probably DxO has in its backlog some PL speedups too, e.g. related to preview caching, which is something to improve indeed. Calling C1 UI ‘more intuitive’ sounds also highly personal, somewhat contradicting my own experience.

CaptureOne is very good for very good input (only).
PhotoLab deals with noise and lens sharpness by far better. C1 has more subtle HSL-like tools, face retouching and skin tones control. PL has better tools for local contrast and overall picture “cleannes” and tonal balance. C1 UI has more customization options, but in general it’s slower to use, e.g. think of mouse-wheel, not to mention clumsy imports. For my type of work I find PL superior to C1 in terms of image quality and ergonomy, so PL is my main workhorse, but I still use C1 for some specifics. For other types of work, e.g. studio, C1 might be a better choice. It’s best to have both :slight_smile:

Summary, which presumably everyone agrees to:
Going “religious” with any software will do you no good.

3 Likes

Like Microsoft, e.g. regarding privacy concerns :slight_smile: :slight_smile: :slight_smile:

Probably like the directory layout designer in this case :slight_smile:
PhotoLab takes a list of all files/directories in its current working directory and reads their properties. Note that directory metadata contains files and subdirectory entries on equal terms. Btw, Windows is quite well known for its long open() processing time (compared say to Linux) and NAS only adds to this latency, not to mention anti-virus overhead.

‘Cancel’ for what? You said that module download has completed within seconds.

That shouldn’t be the case after PL directory change imho, so report it as a bug. How many bytes/file on average were read? Are you sure it was actually reading the files, or was it just inspecting the directory metadata contents?

How do you know that?

That’s very true. On the other hand, it may also hide some race conditions, so it happens sometimes that software runs faster on slower hardware :wink: Both cases should be tested for sanity checks.

After struggling to read your long post, I still have no clear idea what your problem is :frowning: If time permits, I may take a look at your ProcMon dumps if you document ProcMon filter, configuration, actions taken, plus PhotoLab log and relevant directory listings.

That is very fair to say, but before I give the impression I’m a C1 “fanboy”… if I were, I wouldn’t be here.

Fact is I prefer what PhotoLab can do. I prefer its potential.

That’s why I have so much to say about its performance issues, the odd crazy design approach, or the unpleasant business practices from DxO and their poor customer support.

With some work in these areas it could be phenomenal!

2 Likes

I see it exactly this way. PL9 has amazing denoising capabilities. But I’m too old to wait out all these delays because PL9 thinks it needs to read my entire, admittedly huge, image collection.

1 Like

Its a long standing problem that PL reads all drives shown in windows
My laptop has a lot of usb SSDs at diffrent times and PL takes ages to load as a result even if non are attached. My NAS is networked and its checked every time most frustrating that they have never given the operation to have it access a user defined drive

2 Likes

Of course, you can also look downwards. Or what’s the point of your comparison with Microsoft?

You find 5000 entries in one folder strange? The file system itself has no problem with a large number of entries. Problems with excessively long paths, however, are not uncommon.
NTFS has a theoretical maximum of 4 billion entries. Ten thousand entries are absolutely no problem for my computer. But when the program reads the contents of the files and subfolders, that naturally takes a long time. However, there’s no need to read the contents just so I can navigate the directory tree.

It’s nice to be able to build a database. But if I have no control over which data is indexed and which isn’t—and if I have to wait until PL9 has finished—then that’s not ideal.

How do I know what’s irrelevant? It’s not that complicated! If other programs don’t need to read all the images in all the folders to edit a single image, then I can say with certainty: What’s in the other folders is irrelevant for that task.

Thanks for your offer. I’ve sent countless ProcMon logs to support. Unfortunately, to no avail, except for: “Your insights are important to us” - blah blah.

Now my trial period is over. And in the short time I’ve had with PL9.2, I haven’t seen any effort to improve performance.

PL9 is a nice toy for those who want to align their workflow and infrastructure with DXO’s ideas.

That’s why I’ve decided to buy the latest Capture One. The one percent of images where DXO denoising would actually be advantageous are outweighed by the 99 percent where I simply want to work without any lag.