Wow I just tried Google Photos to see what all the fuss is about. Not impressed. Really. I actually think more of the plex implementation than before.
There is a fundamental difference in the Google approach however, more than just AI scoring accuracy. They supply some key interactivity which allows you to shape the tagging process. The plex auto tagging model, on the other hand, is zero user interaction. You get whatever the tagger gives you.
Implementing the interaction and tuning functionality, which I think is key, is a big jump. It will involve significant product engineering and the feature would require a lot of time to mature. Not to mention the cost of the machine time required to produce good similarity scores (it’s not for nothing that they limit processing to X images per time period, it’s a processing cost issue)
Bottom line, I’m not holding my breath on this one, and I wouldn’t really blame plex inc for taking a long time to produce it, if ever. I feel like if I really want this in any predictable timeframe I’ll have to implement it myself.
