Automating Photo Culling

[Ledgem]

I think the only part of photography that I really dislike is photo culling. This is when you download all of your photos and are sorting through them, trying to figure out which ones to keep (and possibly edit), and which ones to delete. In the past, this wasn't so terrible a process. But now, when your camera can easily store tens of thousands of photos on a single card, and when cameras are fast enough that you can easily take a dozen photos of the same thing (even when not in burst-shot mode), it is a painful process. It's no longer a matter of deciding which photos are interesting, but of picking out which one amongst a dozen is the most perfect.

For certain, there's some personal discipline involved here. If you're the type of person to take one shot and then put your camera down, you may not experience this pain so often (and you have my admiration and envy). If you're the type of person to look at the first photo of a series of near-duplicates and figure to yourself that it's good enough and you won't waste your time agonizing over the others, my hat is off to you as well. I aspire to both of those things, but after years, I can say that I am not there.

And are these activities that truly need a human brain? A computer should theoretically be able to detect things like focus issues, blinking eyes, and grouping together photos that are similar. Those things alone wouldn't change my photo-culling world, but they would make it better.

I heard about Canon designing some software to do such a thing, and began to looking into solutions. I'm going to list three here, but as a fair warning, they're all in beta and may not actually be available (so far I've signed up for the beta for all three, and have not been presented with a download link, although on the bright side they do seem to talk about MacOS support):

The first that I came across was Narrative Select. I've been on the waitlist for their free beta since about the last three months. Seems like it has potential, but their marketing scheme is a bit worrisome: you can bump up your place on the waitlist by having more people sign up with your referral code. The link I've posted doesn't include my code, for what it's worth; I am currently somewhere around spot 2,500 on their waitlist (rechecked just now). Seems like there's a lot of marketing hype but not as much people talking about actually using it.

FilterPixel. I came across an actual tutorial of sorts for this one, and it looks promising. Their signup process does ask for an email address, name, and phone number, which I felt was a bit invasive, but here's hoping it works. They also ask if you're using Lightroom - I use Capture One, but these programs can supposedly tag and color outside of the photo libraries (which should be able to import in).

AfterShoot. Also saw some screenshots for this one, I believe from the developer; the signup process asks for your field of photography and for a link to your portfolio (of which I don't really have one - and haven't updated my Flickr in years). The overall functionality seems to be similar to FilterPixel.

It'll be interesting to see where these software projects go, and what the eventual cost is.

Hopefully these links turn out to be of interest and helpful to people, eventually... but until I get one of the betas to start using now (assuming they grant me access), I'm stuck on my own. I used to go through my photos in one long pass, but I will probably start utilizing the multi-pass method for my own sanity.

So how about the rest of you? Are you using any software aids, or is there something else that you do?

EDIT: As I was flipping between the websites, I noticed that both the FilterPixel and AfterShoot websites have some graphics that look extremely similar... and even the images within the programs (based on what they show) look almost identical. Kind of weird, kind of sketchy. Will report back if I get download links from either of them, and if the software seems to work.

EDIT #2: Found another program, Optyx. There's a free trial that handles 100 photos at a time (which is pretty generous), and the full cost is $100. If it works well, I think that's definitely worth the cost. Based on the screenshots, it seems to be more geared around grouping similar photos, but it seems there is an option to automatically cull, although I'm not sure what level of control is given there. I'd give it a go but after just a few hours of waiting my request for the AfterShot beta was approved, so I'll give that a try first. It's exciting to potentially have multiple options with this task!

 

[Ray2]

It’s the price people pay for spraying shots all over the place. I don’t think culling is where improvement is needed.

 

[Ledgem]

I tried out AfterShoot and Optyx tonight. Here are some thoughts.

AfterShoot Beta
AfterShoot's sketchiness at sharing basic artwork with FilterPixel didn't end there. It's an unsigned app, so both Apple's Gatekeeper functionality and Little Snitch threw warnings over that. It didn't seem to access any weird websites, however, so I granted access on a website-by-website basis and ran the program. I threw a little over 3,000 photos at it; it spent about seven minutes "importing" the photos, and then another 30-40 minutes actually running them, during which time the processor and RAM usage seemed to go up appropriately (although not as much as I'd expect for something that I'd imagine to be a fairly intensive task - this was a combination of 20-megapixel files and 50-megapixel files). It offers some nice control over how it rates photos, and I elected to have it use a star rating system, basically assigning one star to blurred photos, photos where a blink was detected, and only coloring when it detected exposure-bracketed shots (because I do HDR here and there). It's focused toward wedding photographers, so it also had an option for kiss detection; I disabled that. My goal would be to have it sort out the really undesirable photos, leaving me to use my basic color system over the star ratings.

In going through the photos, the results seemed promising at first... but quickly fell apart. The color tag was applied pretty aggressively to photos that weren't exposure-bracketed, even though I'd think exposure bracketing should be the easiest thing to detect (if nothing else, basic analysis of the EXIF data should make those stand out). Five- and one-star images seemed appropriately tagged at first glance, but I found many one-star images that were perfectly usable, while there were also a few five-star images that were out of focus (and I even encountered one with both blurring AND blinking). I didn't bother to go through all 3,000 images; the results weren't totally random, but they were inconsistent enough that I don't feel I can rely on it at this time.

To be fair, it's in beta, and when I try to trace this on the internet it seems the beta just came out within the past week. This isn't an overly easy problem to solve, but if they can do it, it'll be very impressive. I'll be keeping my eye on them. If you're comfortable with some of the sketchiness I noted above and in the previous post, go ahead and apply for beta access, and see how it works for you.

Optyx
The only program out of beta and with an actual price listed, I had high hopes. The install requested to install some developer tools, which raised some suspicion on my end. It further gave me a bad impression when I couldn't figure out how to advance; you need to "Create a shoot" but the button was not clickable. After stumbling around for a few minutes, I realized you need to manually enter a name before you can click the button. What ever happened to either warning dialogs, or just assigning a random name based on the current date and time?

I loaded 96 photos in to start with, all 50-megapixel files. It worked quickly. Optyx is a bit more of a managed workflow; to begin with, you choose auto grouping options, which affords you a fair bit of control (choosing how strictly similar images should be; how long between shots images can be grouped by; and exposure bracketing). Auto-grouping would be, by itself, already very useful. You could stop there, or you can move on to the next pane, which is auto-culling. At first I couldn't figure out how to change any settings with auto-culling, but after fiddling around, I found that there's a fair bit of customizability - although it's not always obvious how the ratings and hierarchy will work, when you can assign things like a certain star rating to "best face" and another star rating to "best overall." Auto-culling looks within the groups and culls based on that.

My feelings are a bit mixed here, as well. The grouping seemed to get things right more often than it got things wrong, but even with some of the most lax settings possible, it still tended to either make more groups than there should have been, or it would leave one photo out of a group that should have been included. You can manually override any of these things, but the point here is automation.

RAW viewing reminded me of Photo Mechanic-like speed, which is extremely high praise. The only downside here is that this is a program made for a single monitor, and it showed. You can very quickly toggle between a focused viewing mode and the grid of all photos, but there is no way to see both at the same time. Perhaps more critically, there's no way to view two photos large in a side-by-side manner. This is a blessing and a curse: a blessing in that it would force my indecisive self to just choose a "good enough" photo instead of agonizing over minute differences between two or more photos at 100%; and a curse because I'd probably end up doing that anyway, only then I'd be relying on two programs to complete the culling process.

What about auto cull? I'll have to play with it more. In grouped photos it seemed to pick things decently, although that's purely subjective. Outside of groupings, it would still assign star ratings and "best overall" selection stamps to photos that were mis-focused (including some where nothing was properly in focus). Similar to AfterShoot, it doesn't seem overly reliable, although perhaps part of the problem is the photo grouping isn't even 99% perfect to begin with.

I'll give Optyx another test session or two, but as of now I'm not convinced enough to spend the money on it. It's currently at version 1.2.0, so also in its infancy. I think there's potential, but this may not be the version to jump in at.

EDIT: the talk of auto-grouping reminded me of a program I found and had wanted to try for a while, PhotoSweeper X. Priced at only $10, it originally sold itself as more of a duplicate finder. A very quick play with the demo version shows incredibly fast performance and a promising setup, although it still occasionally breaks photos into more groups than it should (even with the most lax settings). Unfortunately it probably won't be an option for me, as it does not seem capable of displaying my GFX 50S' RAW files (which do not show up in Finder, either). Otherwise, this would probably be my choice - not only based on cost, but on the functionality.

 

[Darmok N Jalad]

As a hobbyist, I don’t feel the need for auto-culling. Cleanup is part of the process where I learn how to be better and to have a critical eye for the content, but it’s also where I can learn what can be saved in post and what its a lost cause. I’ve also heard from others that “sharp” isn’t the only acceptable form of a good photograph, so it would be sad for AI to trash what might be a nice moment because it wasn’t sharp enough.

 

[Clix Pix]

I've been perfectly happy with Photo Mechanic; it's fast and helps me get the job done. I agree with Darmok that while auto-culling may sound tempting, it really is better and a learning tool to cull one's work oneself, even if it is a tedious project or one doesn't always have time to do it.

 

[mollyc]

I cull on my own. Some I can cull immediately upon import. Others sort of linger in LR for a bit but I will do a day where I did a final cull and edit every couple of months. 99% of my photos I export to jpeg and share with family on a monthly basis (but usually I just send out like 3-4 months at one go set up in galleries). When I realize it's been too long since exporting and sharing I just sit down and cull like mad and edit any stragglers.

But I do a lot of nature/macro types of images and I don't think those would work well for auto-culling anyway.

 

[robgendreau]

Nothing here convinces me there's a shortcut.

And there are trust issues. Reminds me of OCR and AI that would summarize text. They can be 90-95% accurate, but what gets lost in the 5% can be killer...think dropping a zero in a contract. Ditto for photos; something unique might be culled when it's exactly the kind of risky shot that actually worked. We've all probably blundered into those.

And in any case, most culling is like speed reading. I can chuck the obvious lens-cap-on and badly out of focus shots so quickly it's not work trying to automate. Similars are hard, but they will be anyway, since slightly different framing, exposures, DOF, etc was used on purpose. Not to mention HDR, focus stacking shots, panos.

But I have found Fast Raw Viewer to be really helpful in culling and ingesting. Not AI or automatic, but well designed just for the task of culling raw images.

And I agree with Darmok too; a large part of every shoot I do is learning from what worked and what didn't, and that enourages me to continue to try to push the envelope.

 

[r.harris1]

I'm also a happy Photo Mechanic user - have been for years. It's fast, hugely customizable and can be as "dumb" or "sophisticated" as I want. I don't do a ton of action shots at a gazillion-billion frames per second so my shoots will have anywhere from 3 or 4 to maybe a few dozen images at most. On the rare occasions I do a gazillion-billion frames per second, all of those shots will linger on my hard drive (Capture One session) for a bit and then I'll choose one or two and whack the rest before adding to a catalog. So it's all by eye and manual for me.

 

[akash.nu]

The thought of auto culling throws a shiver down my spine. My main fear is that a shot might be out of focus but that might be intentional, a shot might look wonky but I might have done it for a reason. Auto culling is too risky for such things.

Generally full automation in any art form is not a good idea. Art by definition is a thing of perspective of individuals, there’s no right or wrong. So the appreciation parameters are different for different people.

 

[Darmok N Jalad]

I’m curious if these programs would delete this photo that I posted to the POTD thread last month.

Blurry for sure, but I still like how it turned out and wouldn’t want it deleted!

 

[mollyc]

I’m curious if these programs would delete this photo that I posted to the POTD thread last month.
View attachment 949584
Blurry for sure, but I still like how it turned out and wouldn’t want it deleted!

Yes I often (okay sometimes) take an OOF image on purpose for any number of reasons. One or two of these rank among my favorite images ever. Or any of my Lensbaby images, which have purposeful blur. Those are also keepers.

 

[akash.nu]

I’m curious if these programs would delete this photo that I posted to the POTD thread last month.
View attachment 949584
Blurry for sure, but I still like how it turned out and wouldn’t want it deleted!

out of focus deletion coming right your way!

 

[angularboy]

Thanks, @Ledgem, I am one of the members at FilterPixel team.
Apologies, if anyone of you signed up for FilterPixel and haven't heard back.
We are sending invite emails for the beta next week.

First of all, I am impressed by the author in putting honest opinions here.
At FilterPixel we are releasing our private beta next week based on first come first serve basis. To answer @Darmok N Jalad, if we would delete this photo which has an intentional blur?
We have worked extremely hard to extract the intentional blur and separating the main subject to check focus.

Unlike other software, we detect focus only on the important subject rather than every subject. Let me give you an example:
in the photo shown below, automatically only the main subject is extracted and checked for focus rather than the others on the background.
PS: We also don't delete any photo but only categorize them into Warnings and Rejections.

We want to boost the productivity of photographers and want them to focus more on creativity.
Always open to thoughts and opinions.

For updates, here are our channels:
Instagram: https://www.instagram.com/filterpixel/
Website: https://filterpixel.com
Email: aayush@filterpixel.com