I2P is completely decentralized, while tor still has some centralization. When you start connecting to i2p your i2p router looks for the IP of a floodfill router, floodfill routers are other routers, just like yours, except they also have the responsibility of sharing other i2p routers they are aware of. Since any router can be a floodfill the network is harder to take down. Obviously i2p has to get the first floodfills ip somehow, and to do that i believe it does the exact same thing as tor, connect to a regular i2p run server and pickup the ips of some floodfills, however unlike tor, once your router becomes aware of other routers it becomes completely decentralized and P2P.

Meanwhile, tor has a centralized list of all the (public, aka not bridges) nodes that your browser uses to make the tunnel.

When you connect to an eepsite (I2Ps version of onionsite) your traffic will leave through an “outgoing tunnel” which consists of 3 (default, this is customizable) other i2p routers and then is passed onto the webservers “incoming tunnel”. Every router has a set of incoming and outgoing tunnels which are used for communication, unlike tor where you have one tunnel that is established with the website and is used for both sending and receiving. When the website receives the traffic, it will respond on one of its “outgoing tunnels”, which consists of a different set of random i2p routers and will send that traffic to one of your “incoming tunnels”.

Because of this, a round trip for your connection consists of 12 nodes total making it far less possible for any participant to be identified, instead of tors 6 (6 for onionsites, 3 for clearweb)

This is the best graphic i can find to explain it since I feel ive not done I2P justice. This video does a much better job at explaining i2p and goes super in-depth.