Staying off Facebook is not enough. Facebook could easily write a web crawler to find off-site photos of people tagged on Facebook. They could even create shadow accounts to track tagged people who don't have Facebook accounts. Even people not tagged could have shadow accounts seeded from Wikipedia or news photos.

Writing web crawlers and astroturfing to get all that data is hard. Facebook would more likely just buy that data straight from the source, since most other tech companies are similarly unscrupulous and happy to sell.

The second information which I see having potential is the EXIF of the pics, the off pixels and other identifying data of the camera (flare, blur, watermark), artificially introduced or not.

Not only it proves who we with whom at the same time. It also links profiles. You can create a whole new identity; but if you didn't throw away all your phones and cameras, then Facebook can link the profiles. Or if you sell something on Craigslist and upload on Snapchat with the same camera: It's tied to your identity.

Was it just to make a point? Anyone can Google "Richard Stallman" to retrieve more than enough training data to train an image classifier.

>Anyone can Google "Richard Stallman" to retrieve more than enough training data to train an image classifier.

Yes but I think it has more to do with tracking him (i.e. on x day he was at y place/event) than merely identifying him.

There are more than enough pictures/videos of Stallman to make a really clean faceprint.