Category Archives: How To

Lightroom 6.0.1 and Facial Recognition Woes

I purchased Lightroom 6 about 4 hours after it was released on April 21st and my computer had been indexing faces from that time almost none to about 4pm Monday May 3rd, so about 13 days later it finished. This was processing 228,000 images, guessing about half had faces in them.

One of the big things Lightroom 6 supposedly brought to the table over Lightroom 5 was performance improvements, but I have not yet seen any evidence of it, but more to the contrary. Though to be honest.. I have been working off my laptop for the last two weeks because my desktop LIGHTROOM (not the computer itself) was so locked up doing facial recognition it was otherwise unusable. So if there are performance improvements in other modules I haven’t particularly noticed too many of them yet.

Thanks to a previous post commenter (JBurroughs) for pointing out that you do not need to be in Facial Recognition mode for it to process images and it seems SIGNIFICANTLY faster when you are in Loupe mode and it indexing as it is not trying to constantly update stacks and order them as they are added.

Lightroom 6 within in the last couple of days distributed a minor update via Adobe update but the version still shows as 6.0.1 (vs. 6.0.0) and didn’t mention any performance improvements. So no help there.

Post Indexing Experiences

  • Post indexing after seeding and initial frustrations of my catalog of 228k RAW images I had about 137,000 faces to identify with the largest stack being about 80 deep and there was less than one screen of image stacks over 50.. so this is going to take me quite a while still to work that number down especially with the performance issues still occurring in the application.
  • In particularly facial recognition grid view is still ridiculously slow, simply selecting a row of images can take 5 minutes while you wait for Lightroom.. no CPU or memory spikes while it does it either.
  • When you go into people view, it still attempts to do “Finding Similar Faces”, even though indexing is completed and if you pause Facial Recognition. This causes significant pauses and delays in processing if you try to do anything before it is done with this “Finding Similar Faces” process…wait for it… it also creates a TON of noise generating 40,568 “alternative” suggestions not listed in the main view.

So my TIPS for proceeding..

  • I find the best way to tackle the now indexed photos is going back and doing directories of about a 1000 images or less, and even then sometimes has a bit of lag but mostly is barely notable, if I get over 2000 images I start to notice it even with large stacks.
  • I then , if it is all of a single or small number individual (such as from a portrait shoot) I select all images in the directory and then go and deselect any images not them or false positive facial recognitions (such as bushes) and ignore ALL suggestions if I know who they are. I then type their name and let auto-complete address the name if it is an indexed person and I basically override all suggestions. (So basically for the most part completely devaluing facial recognition and recopying the keywords I already had in place).
  • I then go to small groups and repeat the process for them just with more selecting bulk selects.
  • I then proceed to filter out the biggest bang for the buck crowd directories where I do not care. This includes people’s faces in airshows, street photography, weddings (eliminating all but the wedding party), concerts, etc. Going in and getting rid of facial recognition on these directories in bulk. This will HUGELY reduce your number of phases to identify.. what is not clear yet is if you can later go back and “re-process” the directory for facial recognition without having to also manually identify each one.
  • I then go to larger top level directories of about 1000-2000 images.
  • I take on everything else.
  • MOST IMPORTANTLY, every so often go back and check your named people to make sure the faces at least match and you do not have some bad seedings that delays in Lightroom didn’t accidentally result in people getting labeled incorrectly. Do this by clicking on the named people and then when you find someone who is not that person click on them and give them the proper name just like in the main view.

At the current rate I am estimating I will be retagging images for months, maybe they will have an update by the time I get done but not seeing how it will work into my regular workflow and maybe just be a background idle process activity.

Other observations:

  • My catalog went from about 3.3 gigs optimized to just over 4 gigs.. so about about a 20% increase in size by cataloging the faces within my catalog. (Note: I didn’t look at my catalog size before hand, but compared current vs. last backup so this number is not precise, just an estimate)
  • There have been many posts about people grumbling about it identifying faces in magazines or billboards in the background.. I honestly like that it can do that as it helps with identifying people in a crowd.. though it needs to be better. I find it only gets about 70%-90% of the people in a wedding formal photo’s face and misses the others, even the ones completely standing out (like the bride and groom).
  • I have noticed it tends to have high false positives on some women’s floral dresses, especially if there is a grid pattern and the flowers are within the grid pattern.
  • I REALLY wish I could disable facial recognition on a particular directory. A have several of directories of event and street photography where there are large crowds of people and it picks up dozens of faces to identify of random people I will never know (or probably care) what their name was. Yes I can reject selection and then delete facial recognition markers for those images, but that creates a lot more work for me.
  • I have noticed it commonly identifies double round edge numbers, especially “22”, as a face. I have seen at least 2 unique cases of this where 22 was on bleachers in one case and 22 was randomly in the background on something else.  After I saw it a couple times I took a screenshot for an example, in this particular example it came up with at least 4 different names for the front of a treadmill in a gym from an fitness shoot. I did go back just to be sure and checked the 4 names lists and NONE of them also had been accidentally trained to recognize “22” as that person too.

 

Also posted in Sharing Experience

Lightroom 6 – Facial Recognition Performance Issues and some tips to work around it.

Earlier this week I was bait and switched by Adobe to upgrade my Lightroom 5.7 (per a notice in Lightroom 5).. and then it took me to Buy Lightroom 6 when is use the ‘download’ link.. I was more than happy to buy Lightroom 6 (as I am not sold yet on being locked out of my photos if I let me subscription lapse) and I immediately dove into their Facial recognition feature with the hopes it would help me clean-up some of my catalog for untagged images.

I encountered some issues described below as well as developed some useful knowledge to share that hopefully will reduce some of the frustration for others at the bottom of this post.

I watched the Lightroom 6 Facial Recognition introduction tutorial which leaves out a lot of the bulk editing and says basically let it loose on your whole catalog.. NOT recommended (see below) It got me started but also put me down a path of frustration.

So after about 30 minutes of watching Lightroom slow down and it becoming less and less usable while also tagging everyone as the same person which look no where a like. (see below)

I can ensure you each of these individuals are not “Courtney” when you let it loose on the entire catalog. Build a small reference library first.

So I started out with a couple small portrait galleries of 50-200 individuals each 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 after having about 300 reference individuals.

Then I started it out on larger directories of 20000 images and this is where it really became annoying and almost unusable.

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

  • [PERFORMANCE] Lightroom 6 Facial Recognition goes to a ridiculous crawl after about 2000 “unnamed people” stacks to be confirmed (i.e. 2000-2300 in a couple hours, 800-1200 in the next 12 hours).
  • [PERFORMANCE] Lightroom 6 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.
  • [PERFORMANCE] Once I select and click confirm it takes up to 2 minutes to update the “to be confirmed” list again.
  • [PERFORMANCE/BUG] Usually I can perform a couple operations in between refreshes, other times Lightroom looses them and incorrectly chooses the wrong image (selection vs. what is being displayed).
  • [BUG!!!] Command-Click / Control-Click selection sometimes does not work for individual images.  This could be when trying to select a single image or after you have already selected many and you are trying to add an additional image. The BUG sometimes follows the stack for many refreshes, but sometimes also not. The confirm and reject buttons when done individually work and they can be part of a shift-select group.
  • [BUG!!!] 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”  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 6 application itself.

I have reported the bugs and issues to the Adobe forums.

Here are a few tips that I have figured out to ease the pain of using the new feature and once the issues have been resolve will still be useful in the future for using the Lightroom 6 facial recognition feature.

  • Only work in large groups selections when actually performing an action.
    • Select 10-50 image groups at once
    • Use the control select, not just the shift select.
  • Pause Address detection / lookup; unless you geocoded your images this is a waste of time and also can be done as a separate operation.
  • Pause the Address Verification and Face Detection when you need to do things

  • You can select multiple images at once for more than one person and accept or reject in bulk (as long as the same operation). This took some experimentation and risk to realize it behaved this way.

Example 1:  If I click accept on “Anya” (where the red arrow is) it will not assume everyone is “Anya” but the name suggested for the selected box.

Example 2: If I click reject on “Anya” it will return all selected images to a “?” status and Lightroom 6 or Lightroom will make an alternative suggestion on the next update. (As these are all Melissa C.)

Example 3: Being I know all images in all two rows are Melissa C. , I can select all images and type “Melissa C.” and no additional clicking is required. It will update all images to Melissa C.

  • Wait at least 10-20 seconds between actions to let Lightroom 6 catchup and perform any updates or I have had a few times Lightroom take my subsequent actions occur on a resorted and updated image selection which I did not intend.
  • If you know you have a lot of images of a single image in a Catalog, once ID go into that particular individual by clicking on them and get them out of the way by letting Lightroom focus on finding similar individuals. Once you do tis for a few high frequency individuals it will make a noticeable difference and also greatly decrease false positives for that individual in the future.
  • Don’t waste your time chasing Lightroom while it is cataloging trying to update unless one of two things are occurring.. (Simply, if it IDs them correctly.. leave them alone as it will group them together for confirmation eventually leading to less clicking… it took me a long time before I realized I just needed to wait and check in on occasion. )
    • #1 It is identifying the same individual in a large number of photos and does not have a suggestion.
    • #2 It is incorrectly classifying people and you think you can correct future suggestions on a high frequency individual.
  • Unless you use other programs to edit your images and need the XMP sidecar files to recognize Lightroom changes, you can speed up the process a little bit by disabling writing XMP data in the Catalog settings, even if temporarily. Note though if you do this any facial recognition data will not be written to the sidecar.
Uncheck

Uncheck “Automatically write changes into XMP”

 

Also posted in Sharing Experience Tagged , |