Hacker News Reader

Comment by 27182818284 | Hacker News Reader

Wait, what. At roughly the 1:34 mark, the voice pulls up photos of her kid with a pumpkin. Do you have to tag your photos with "pumpkin" first or is the recent "Choose the picture of waffles" in Recaptcha fueling that?

View on HN

No, they are automatically classifying photos. If you have photos in Google Photos, you can try it out with words such as "forest", or "dog", or "grill". It's pretty cool and scary at the same time :-)

I did not know they were doing anything other than identifying faces that you've already tagged. Grabbing the pumpkin, whatever out of the photo is just wonderful.

It's even better than that. It can recognize landmarks. I can ask it "Show me pictures of myself in Paris" and it finds 25-year-old scans that I uploaded that haven't got any kind of geographic data in the EXIF.

You don't even need to tag faces. It finds all the faces that have been in your photos and picks one for each to be the search "face" for that person.

The technology underpinning this stuff has been brewing / percolating / developing for years [https://play.google.com/store/apps/details?id=com.google.and...]

Automatic photos classification is just the most visible iteration of the tech.

It's pretty terrific. You can search for things like "wedding" or "beach" or "birthday" or "christmas", people's names, "march 2008", "kyoto", etc, and it uses everything at its disposal: the image itself, the EXIF data, geotagging, the date it was taken, the filename, the album name...

The '"Choose the picture of waffles" in Recaptcha' actually is training a neural network, which then is used to tag your photos.

Reply on news.ycombinator.com