Home▸Machine Learning & Neural Networks▸Capsule Networks: Preserving Spatial Hierarchies

🧭 Capsule Networks: Preserving Spatial Hierarchies

Explore how Hinton's capsule networks replace max-pooling with vectors of neurons and dynamic routing-by-agreement, letting a network preserve precise pose and part-whole relationships that ordinary CNNs throw away.

Machine Learning & Neural Networks3DModerate60 FPS

A layered capsule network where lower-level part capsules (rendered as small oriented vectors) send weighted predictions upward to higher-level object capsules, with connecting lines that thicken and brighten over routing iterations as agreeing predictions cluster together and disagreeing ones fade out.

🔬 What It Demonstrates

🎮 How to Use

Choose an input pose or scramble the part arrangement, then step or auto-play the routing iterations to watch coupling coefficients shift toward whichever higher-level capsule the lower-level votes agree on, and rotate the viewpoint to see how pose vectors update.

💡 Did You Know?

In Hinton's original 2017 capsule network paper, just three routing iterations were typically enough for the coupling coefficients between lower and higher-level capsules to settle into a stable agreement.

⚙ Under the hood

Watch lower-level part capsules vote for higher-level object capsules, with agreeing votes growing stronger over several routing iterations.

capsule networkdynamic routingpose estimationmachine-learningcomputer visionhinton

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install