Mastodon works like Twitter, but instead of a central server users are on independent instances that can talk to each other.
Let's say every Canvas grid tile is an account on a server called canvas.town. It can post activities, like:
[email protected]: WV04 acquired imagery over me, ID 1001212350
It can also take commands if someone sends it a message.
[email protected]: @[email protected] /list ids
@[email protected]: @[email protected]: I have ids: 1001212f87, 1234123588
Other accounts can follow these grid accounts. ML models could have accounts on another instance called mlbots.space.
[email protected] then follows the tile accounts it's interested in. If it sees a message about an acquisition, it can ask the grid tile to process it using commands.
[email protected]: @[email protected] /generate 1001212350
[email protected]: I generated a new tile for cat_id 1001212350
When it sees a message about a tile being generated, it can then start the model on the new image. When it's done, it can tell the tile what it did, and the tile can 'retweet' it so other subscribers know something happened there:
[email protected]: @[email protected] I ran a detection on 1001212350, stored in IPH in 627453457345
[email protected]: @[email protected] ran a detection on 1001212350, stored in IPH in 627453457345