Photo analysis program to suggest keepers?

A little background: I have several tens of thousands of photos on my computer and backup drives, and would like to start to go through the backlog to eliminate unnecessary duplicates, missed shots, etc. I'm looking for a program that might help with an initial cull of clearly bad photos (out of focus, extremely under- or over-exposed, etc). The technology clearly exists, given my Pixel phone's ability to take multiple photos and suggest the best of the bunch, but may be proprietary. After several hours of googling, I can't seem to find anything that's readily available.

An extra bonus would be something akin the the Pixel analysis, which could analyze multiple similar photos and suggest the best one. Quite a few of my photos are in bursts, where there may be 5-10 very similar photos in a row. Some analytical way to identify the keeper from each bunch would be a huge boon. There seem to be some programs that can identify exact duplicates, but those seem fairly cumbersome and might not help with my issue of separate photos from a burst.

Obviously, the ultimate answer will be to just start going through them all a little at a time myself, but it would be nice if there were something out there that could take an initial pass and help limit the total set I'm reviewing. Any ideas?

I don't know of any such package I'm afraid. The main reason is that it is very hard for a program to understand art and creativity. You mobile phone software almost certainly works on the basis of a few very simple criteria. "Are their eyes open", "are they smiling", "Is the landscape generally in focus", "is the thing that was being focused on, in focus?"

Unfortunately when you move onto more complex images (of the sort taken with DSLRs) where there may be intentionally out of focus areas due to DoF or motion blur. The software simply can't understand the creative intent for a particular image and thus can't evaluate it. They are also not good at pattern recognition, which is why Capture/Recapture software fools bots but not humans.

Human's, on the other hand, are very good at pattern recognition. It is how we can recognise someone walking two hundred yards down the road, even though we can't actually see them clearly/their facial details. Yes, culling is a pain in the arse but doesn't take too long if you trust your instincts. The key is to try and avoid "decision paralysis". Here's what I do....

Load a set of images in Lightroom and display the first one (Loupe view). If it is OK I press 1 (star rating), if it looks good/great press 3 (star rating), if it is out of focus/bad press X and then instantly on to the next one by pressing the arrow key. I DON'T look at each image in detail. This is an initial pass to get rid of the obviously bad images. If I have a burst of similar images I don't try to pick the best one (choose 1 from 6 = decision paralysis). Instead I pick the worst two/three. Flip back n forth through the burst a couple of times quickly, mark the worst ones with an X and then move on. Any difficult decisions - I postpone them. Keep moving forward, don't get bogged down.

When I get to the end of the first pass I Delete all the Xs.

Now I just have the 1 star and 3 star images.... so I filter to only show the 3 star images and I look through to see if I have enough images for my client/my own usage. If I do then I don't need to look at the 1 star images (I don't generally delete them). If I have a burst it is now a burst of three instead of six. It is much easier to choose the best of three compared to the best of six. Now you can return to the difficult images to study them, without being overburdened by the sheer mass of images you need to sort.

If I don't have enough 3s then look at the 1s. Are there some that may be a 2 star or even a 3 star with a judicious crop. If so do that until I have enough pictures. Job done - send to client/post for family to look at/whatever.

I now leave the remaining images for a week or two. Go and do something else/cull a different batch. I let myself get a little space between the images and the pressure of culling. I return to them with no need to do anything in particular. I can just look at them in a relaxed manner to see if there are any hidden gems that I overlooked (there often are). If so I process those and add them to my 3 star batch.

Google is using machine learning algorithms (AI) to select the best picture of a series. They most likely are not choosing features to qualify pictures (this is considered bad practice in ML circles) but just use their huge database to train an AI what good pictures usually look like. Software for these purposes are freely available and not too difficult to use. You can use them to easily train them what YOU personally consider 'good'. The main challenge is, that you need to train them. The larger the diversity of pictures the more training data you need. And this is exactly what you try to avoid. Google and Facebook have hundred of millions volunteers to do that for them. I am quite confident, that anybody who has a well trained AI will keep this a closely guarded application....

If you shoot raw, then fast raw viewer is what you want to use. It will not choose anything for you, but it will render your raw files, with a raw histogram and over/under exposure indicators, shadow and highlight inspection, focus peaking all in real time. It makes plowing through a set of raw files a piece of cake.

Kirk

If you shoot with Canon DPP is a great culling tool. Quick Check - Full Screen mode. Just hit the X key for unwanted flies then Edit - Rating - Select Rejected Images Only followed by File - Move to Trash. DPP's downsizing algorithm for 'fit to screen' display is very good - contrasty which creates the illusion of sharpness.