Lightroom 6 / CC2015 - Facial Recognition Terribly Slow, not using all of CPU or GPU, still keeps going when paused.

[crsouser]

This is actually a couple of issue but wondering if others are experiencing it and worked through it.

I am using a Dual Xeon CPU 3.2 ghz Mac Pro, Catalog on SSD, Images files on Mirrored Pair, Dedicated GPU (max I can install in my version of Mac Pro) and hardware acceleration enabled in Lightroom.

I watched the Lightroom 6 Facial Recognition tutorial which leaves out a lot of the bulk editing and says basically let it loose on your whole catalog.. NOT recommended.

I started out with a couple small portrait galleries that identified a couple hundred total people to seed facial recognition so it didn't suggest everyone is the first person I confirmed (which it will do otherwise). I have also optimized my catalog multiple times.

I have encountered the following serious performance issues and bugs with Facial Recognition.:

• Lightroom Facial Recognition goes to a ridiculous crawl after about 2000 images to be confirmed. (i.e. 2000-2300 in a couple hours, 800-1200 in the next 12 hours)
• Lightroom becomes largely unresponsive after having a fair number of images to be confirmed, even after pausing Address and Facial Recognition. So even selecting 4 rows of images can take 5 minutes with several long pauses.
• Once I select and click confirm it takes up to 2 minutes to update the "to be confirmed" list again.
• When I click on an individual at the top of the page, pause facial recognition and address lookup it still continues to "Look for similar faces" [BUG!!!!!!!!] even though all I want to do is just confirm some individuals more quickly in bulk with the images already identified.. not continue to look for more as a work around for the painfully slow responsiveness of the module.

The odd part is that with all of the performance issues Lightroom will not use more than 20-30% of my two Xeon CPUs, barely touches my GPU (<10% CPU, 30% memory), my and no more than 35% of my memory. Computer Temps are also barely above startup temperatures and 15-25 degrees cooler than when I run other applications which will consume my entire CPU and memory if I let it. I have explored Lightroom's settings but seen nothing further I can configure to speed it all up.  I have also attempted the operation on images on the SSD, my drobo (known to be slow), an independent fast disk I have, and a pair of raided disks and have the same issues.

I will also note that all of my other applications seem to continue to operate just fine.. the slowness seems to be contained to the Lightroom application itself.

Lightroom version: 6.0 [1014445]

Operating system: Mac OS 10 Version: 10.10 [3]

Application architecture: x64

Logical processor count: 8

Processor speed: 3.2 GHz

Built-in memory: 18,432.0 MB

Real memory available to Lightroom: 18,432.0 MB

Real memory used by Lightroom: 5,537.5 MB (30.0%)

Virtual memory used by Lightroom: 32,240.6 MB

Memory cache size: 4,342.0 MB

Maximum thread count used by Camera Raw: 8

Camera Raw SIMD optimization: SSE2

Displays: 1) 2048x1152

Graphics Processor Info:

NVIDIA GeForce GTX 285 OpenGL Engine

Check OpenGL support: Passed

Vendor: NVIDIA Corporation

Version: 3.3 NVIDIA-10.0.31 310.90.10.05b12

Renderer: NVIDIA GeForce GTX 285 OpenGL Engine

LanguageVersion: 3.30

Application folder: /Applications/Adobe Lightroom

Library Path: /Users/DryClean/Documents/Lightroom_Catalog/MyCat_LR6.lrcat

Settings Folder: /Users/DryClean/Library/Application Support/Adobe/Lightroom

Anyone have any suggestions?

[RyanMatthewBarrett]

Complete.  Not sure when.  I ended up with 13k faces from 53k images.  I think there are about 2000 of my wife that it couldn't stack together. 

[Kim_Letkeman]

It's like it was written by high school kids. No separation of GUI and worker threads, no proper status on each face so that you can click ahead and feel safe that you are not doing random stuff. Terrible use of the system yet the ability, no matter how low you set the thread priority, to cripple the mouse. All in all, probably the worst implementation in any of the many modern and very expensive applications I use. That team should feel ashamed.

Note: 8 core 8150, separate striped cache and scratch disks, large and fast image disk. 24GB RAM. SSD boot drive. About 120k images in the folders. Can't leave it running as it bogs down. May have to give up on tagging as I don't have the time to baby sit this piece of junk.

[barnao]

I think you kinda nailed it regarding the coding practice... it's astonishing.  I get the roots of the application (older architecture) but there have been years that have passed without any significant REAL improvement.   It's inexplicable why so many of the core routines (not just Facial Recognition) use 1 or 2 of my available 5820k/4GZ/16GB/SSD Boot/Raid0 system.   (Facial recognition only being the latest).  I'm not sure if this app is just so nitch (no one using?) or just lack of focus on the team or whatever.  I work on development of large scale applications for a large enterprise with thousands of applications in our portfolio.   Where i work this would be grounds for team swap.

[RyanMatthewBarrett]

I would feel very foolish if I bought an 8+ core Mac pro, and it hung on facial recognition.

8 'core' AMD worked fine, but it clearly wasn't optimized by Adobe.

[M.D.L.]

Kim Letkeman wrote:

No separation of GUI and worker threads, no proper status on each face so that you can click ahead and feel safe that you are not doing random stuff.

I think this is the key problem I am concerned about.  I can't imagine any programmer worth their salt linking the GUI to the worker threads the way LR 6 seems to.  I can't say how many times I've clicked on something and then LR has registered my click later and done something opposite of what I want.  The current implementation is abysmally slow and buggy, and I can't believe they released it like this.  Did they not test it on a catalog with 10s of thousands of photos?  I'm really disappointed with this feature, which I was greatly anticipating before I actually tried to use it.  I've been a LR customer since version 1, and this is by far the worst release!

[knipsefritze]

‌Also, LR 6 was running very unstable and crashed quite often while having the use of the graphic card enabled. Once I turned that off in the prefs, LR was running stable again. Not a very good implementation of that feature either!

[RyanMatthewBarrett]

I haven't noticed any problems with GPU acceleration turned on. But it's not even noticeably faster or different, come to think of it.

[MM-Webconsulting]

ADOBY, WE NEED A SOLUTION!!! NOW!

[hhelmbold]

Just adding my voice to this post as I am experiencing the same issue that it cannot be switched off or paused. Lightroom carries on! This is ridiculous!

I love the facial recognition idea and I work with individual catalogues per shoot, so I can decide which catalogue to index and which not, but Lightroom forces you to have it on every single catalogue.

[StuartBerman]

I agree with everyone here. This feature is a great idea but no way is it ready for launch. The biggest issue is that even when you get through all the wrong identifications you find it missed half the faces. Even ones that are so plainly obvious are missed.

Anyone upgrading from LR5 will be horribly disappointed.

I really hope they will release a better version as this one is a shambles.

[MM-Webconsulting]

There was an update on CC 2015.2 yesterday and the problem is still not solved!

ADOBE, why?

[mj33641480]

Just want to add a +1 to LR being excruciatingly slow. I came from Picasa because I had heard alot of great things about LR, and just looking through LR it seems functionally more mature than Picasa, but good god, I am not going to wait 5 to 10 seconds between clicks just for some extra functionality.

I will check back in next release to see how things are progressing.

[thwald]

First off, I agree with everyone here so far in that the kind of performance we see in this product is appalling. This is a professional product from a company that should be familiar with its users. This is not a computer problem that is too hard to fix, as others (Picasa for example) have solved many years ago.

I have about 60,000 photos in  my library. It took my Macbook Pro (Core i7, 2.3 Ghz) a full day and a half to find and index all the faces. It started out pretty fast but once the number of named faces reached about 1000, it slowed to a crawl. at about 5000 faces, it was no longer useable.

My take on why the UI slows down is because it is just the UI that is slow, not the worker threads (if there are any). Lightroom seems to update the suggested name of every single photo in your library at every click, including (somehow) those not displayed on the screen. Therein lies the  problem and also a workaround.

A Workaround:

I discovered the following by accident when I worked with faces form the "Previous Import" smart collection.

In your folders view, select only a few folders to limit the number of displayed photos to several thousand. Now I am not sure how reasonable this is for everyone who uses Lightroom face recognition (eg, some shoots may include more than several thousand photos), but...

THIS DOES WORK for me. The UI becomes lightening fast when renaming photos. I would be interested in knowing if this helps anyone else.

Good luck!

[Guest]

@thwald your workaround fixed the issue for me..thanks! 

What I did is had Lightroom crunch through my entire library detecting faces which took 20 hours.

Now I just drill down to the year folder I want to work on and the UI is lightning fast applying tags.

If I work on the top level folder for faces the UI crawls.

Working from the year level is what I want to do anyway.  It's a reasonable workaround for me!

[Mixelmagic]

In my case face recognition became bearably fast when I disabled GPU acceleration. I realised mine (512Mb 6770M) was way below the recommended spec for that feature.

With GPU acceleration enabled every single click, or face assigned caused a really long wait. Now there are short waits but nowhere near as inexplicably long, so that's something. 😕

This really needs fixing. Photos, Aperture, and Google Photos can all do this stuff with a responsive GUI. This should be a premium implementation.

[marcin65906372]

+1 - I have exactly the same problem. PC with a very good spec (i7, 16GB RAM, SSD). Performance when tagging people (answering to the proposition: yes /no) is extremely slow. CPU is utilized at 10% maybe. Disk usage is at 2%. Not sure what's wrong, but seems like some crappy issue in Lightroom app. Looking forward for resolution - Picasa was way more better & responsive in this area.

[Pat Wibbeler]

I have a 65k photo catalog and found some of the same performance issues described here with performing operations on large sets of photos. Confirming takes many minutes to re-search for "similar" photos in the People Viewer and large selections make the browser less than responsive.

I found that the experience was MUCH better if I constrained my working set to 1000 photos or fewer by choosing a subset of my files in the Library module before entering the "People" view. When I increased that to 5000 photos, it was almost manageable as long as I didn't try to "confirm" too many photos at once. I'm just going to work 1000 photos at a time. I am sure that this number will vary depending on system performance. Here is mine for reference (a roughly 5 year old, but highly spec'd PC at the time):

• Intel Core i7-2600k - 3.4GHz
• 16 GB Ram
• Catalog on SSD, photos on regular spinning disk

I am an Adobe employee, but I don't work on Lightroom so please don't consider this an "official" response.

[hungchi]

I'm also seeing major performance issues in the people view. removing or confirming a face region which results in a not responding state.

Watching the system performance I see 18% CPU use and very high disk uses.

Seems like the performance issues are mostly related to a lack if optimization or lack of GPGPU processing resulting in higher CPU use and disk related performance issues. (I'm planning on doing some additional SSD drive tests)

I have followed the recent performance recommendation to drastically increase the camera raw cache I've set my cache to 100GB

Here are my system specs

Lightroom version: CC 2015.6 [ 1078672 ]

License: Creative Cloud

Operating system: Windows 10

Version: 10.0

Application architecture: x64

System architecture: x64

Logical processor count: 8

Processor speed: 3.3 GHz

Built-in memory: 32717.7 MB

Real memory available to Lightroom: 32717.7 MB

Real memory used by Lightroom: 2913.0 MB (8.9%)

Virtual memory used by Lightroom: 2994.3 MB

Memory cache size: 6380.9 MB

Maximum thread count used by Camera Raw: 8

Camera Raw SIMD optimization: SSE2,AVX,AVX2

System DPI setting: 96 DPI

Desktop composition enabled: Yes

Displays: 1) 1920x1200, 2) 1920x1200, 3) 1600x1200

Input types: Multitouch: No, Integrated touch: No, Integrated pen: No, External touch: No, External pen: No, Keyboard: No

Graphics Processor Info:

GeForce GTX 1070/PCIe/SSE2

Check OpenGL support: Passed

Vendor: NVIDIA Corporation

Version: 3.3.0 NVIDIA 368.81

Renderer: GeForce GTX 1070/PCIe/SSE2

LanguageVersion: 3.30 NVIDIA via Cg compiler

Application folder: C:\Program Files\Adobe\Adobe Lightroom (SSD drive)

Library Path: Z:\Pictures\Lightroom\LR4-CATALOG\LR4-CATALOG-2-2.lrcat (7200RPM HDD)

Settings Folder: C:\Users\[user name]\AppData\Roaming\Adobe\Lightroom (SSD drive)

Images are also on Two other internal HDD Drives (4 drives in the system)

[hungchi]

Debating what an appropriate preview cache size is. Since I'm hearing a lot that larger is better I'm following that logic.
I'm raising my preview cache size from 100Gb to 250Gb. Here is how I determined my new starting point.

Standard preview size = 1920px

[ (preview size in pixels) * (preview size in pixels) / ( (bytes to killobytes) / (killobytes to megabytes) / (megabytes to Gigabytes) ) ] * (images in catalog)

[(1920*1920) / (3*1024) ] * 62300 = 213.89 Gb (rounding up to 250Gb)

I'm hoping that will help

[hungchi]

a dedicated SSD for the catalog did not resolve the issue

[DdeGannes]

Facial recognition is of course a useful tool for many individuals however its inclusion into Lightroom, which (for me) primarily software to process and render large volumes of RAW image files from Digital Cameras, has created issues that slow down the processes in Lightroom. The program has become bloated and there are some features like this one should be removed to ensure that the core functions function as they should.

[hungchi]

The simplest way to deal with the slowness of Facial Recognition is to avoid working in the "all images" or other folders/collections that have more than 5000 images. With my catalog of 70,000 images it's a nightmare trying to work with Facial Recognition. If I stick to small image groups (working year by year, or project by project) it's reasonably fast and functional.

IMO the majority of performance issues are all in I/O with the disks, rather than read/write speed it's in the simultaneous access of many files all reading information. It get's expensive to deal with limits on input/output operations can be performed per-second (IOPS)....(Multidisk Arrays – More disks in the array mean greater IOPS.)

[TayUK]

I have the same issues plus a couple of others to boot.

1. The facial recog part of the software is 'okay'ish'  It kind of does what it says on the can..just about.

Thats where the praise ends..!!

2. Painfully slow, tedious and downright awful..

I've got dual mirrored 12TB Thunderbolt storage, I/O's are not the issue, its bad coding.  My storage remains lightning fast despite Lightroom doing its stuff,  in fact I was able to run several backups without LR noticing that anything was happening.

3. Saves or lack thereof..

If you feel you must use this feature, do not commit too much effort into it.  I spent the better part of 12 hrs over 2 days sorting, arranging, identifying, correcting,  only for LR to lose at least 5000 named people out of 7500. 

Thank christ I do wildlife mostly.  65k images processed with approx 7500 faces (or objects that LR 'thinks' is a face) to lose 5000 or so was a bitter pill I do not ever want to repeat. 

I'll be exporting my 'faces' to Apple Photos because that at least remembers faces !!  the workflow sucks but for faces it works.

The fact there has been do formal response to this issue speaks volumes to me.  If Lightroom/Photoshop wasnt so key in my workflow  I wouldn't be using it ay all.  

Not only do we have to be lied about security issues, we get the hard sell at the end of each year of subs.     

Just my 2 cents worth.

[Kongulu]

I am amazed how poor the Lightroom facial recognition works relative to the Picasa facial recognition. This issue is even more troubling in that it is now 2017 and the Lightroom facial recognition routinely identifies non face objects as faces.

I get the sense that the Lightroom developers consider face recognition a checklist feature mostly for the benefit of non-professional photographers and therefore not a core feature worthy of serious developer time.

My opinion as a non-professional pro-summer is that Lightroom really needs to address the following issues in order to retain its position over the next 10 years.

+ Speed up everything, especially when working with large catalogs. I have over 100,000 photos and working with Lightroom is tedious. It turns out that after a certain point no matter how much hardware you devote to Lightroom it only uses a fraction of the resources. I have purchased the fastest Arrays, SSD's, CPU's and goobs of the fastest memory along with top end Nvidia video cards - and I still just sit and wait for Lightroom to process my facial recognition and management tasks. It is embarrassing in 2017. I can only surmise that Abobe has no real interest in Facial Recognition.

+ Make better use of top end hardware. This means enabling those users with the best hardware to take advantage of their investment. I invest in top end hardware to save time, but I don't get the sense that Abobe really sees saving my time as a market opportunity.

+ Focus on making Facial recognition fast, accurate and easy to use - the payoff will be immense - and if done well the professionals will be impressed and start to actually use Lightroom facial recognition.

[juliusm61723319]

IMO, this capability is not ready for use at all. It completely ignores people that are viewed from the side (profile), or wearing sunglasses. Its accuracy is completely inadequate and it continues to do very rough mistakes (mixing a guy with beard with a young girl) after getting thousands of images to learn on. In short - it is useless. It is heavily behind the industry state of the art. It can't be compared with long existing face recognition capabilities of Google or Facebook. It is a shame that so professional software company releases such poorly designed feature.