Foundation4D hits 18.1 mIoU on OpenOccupancy, best among compared methods
lealtaixe · x · 2026-10-29
Foundation4D's semantic occupancy branch describes which parts of 3D space are occupied and what is there. After benchmark-specific fine-tuning, it reaches 18.1 mIoU on OpenOccupancy, the highest among methods compared in the paper.
This adds to the framework's other results in recovering depth and camera pose from multi-camera video.
More from Embodied
- GPT-6 Sol beats Gemini 3.8 Flash 1012–972 on Physical Agent Capability Index — qinzytech · 2026-10-29
- Mila's open-source Harpia uses drones + AI to survey hundreds of tree crowns in a day — Mila_Quebec · 2026-10-29
- "Teaching Robots to Be Human Might Be Humanity's Last Job" — and It Looks Hilarious — VraserX · 2026-10-29
- Japan's MW plans ceiling-rail housework robots built into homes, targeting 2028 sales — chris_j_paxton · 2026-10-29
- SO-101 Bench: a $120 real2sim2real robot benchmark where sim rankings transfer to real hardware — chris_j_paxton · 2026-10-29
- Hiro's Origin robot brings automation to small businesses at $9,950 — chris_j_paxton · 2026-10-29