Slot-JEPA: Object-Structured Prediction Interfaces for Latent World Models
What this work does
Slot-JEPA reorganizes dense visual features into stable semantic slots and changing spatial masks. Transition memory and transport-based goal costs make predicted action consequences more useful for visual planning.
THE RESEARCH CONVERSATION
Discussion & reactions
Share a question, an insight, or a replication experience. Sign in with GitHub to join the conversation.