Capsule Compression refers to techniques for reducing the dimension or number of capsules in Capsule Networks to improve efficiency and scalability.

Methods such as weighted sums, dimensionality reduction, or clustering are often used to combine or eliminate redundant or less important capsules.

The goal is to reduce the complexity of the model without significantly impairing the ability to recognize pose and activation information.

Capsule Compression thus supports the application of Capsule Networks in resource-constrained environments and speeds up training and inference.