- VIS-OCCANY // W7
- Python
- ANIMA
VIS-OCCANY
3D Occupancy Prediction That Works Where It Has Never Been
Autonomous vehicles and robotic platforms deployed to new cities, new terrain types, or new operational theaters fail on distribution shift. 3D occupancy models trained in European urban environments misread Middle Eastern cities; models trained on highways have no concept of unstructured intersections. Retraining for every new deployment theater is not operationally viable. VIS-OCCANY (codename: Izanagi) implements OccAny (CVPR 2026, arXiv:2603.23502), a generalized unconstrained 3D occupancy prediction model designed explicitly for zero-shot generalization.
MODULE STATUS: PRODUCTIONZero-Shot Transfer
PROD
- DIVISION
- GENERIC
- WAVE
- W7
- DOMAIN
- GENERAL
- WAVE 7 // ANIMA SUITE
- FOUNDATION — VIS-OCCANY
THE PROBLEM WE SOLVE
Autonomous vehicles and robotic platforms deployed to new cities, new terrain types, or new operational theaters fail on distribution shift. 3D occupancy models trained in European urban environments misread Middle Eastern cities; models trained on highways have no concept of unstructured intersections. Retraining for every new deployment theater is not operationally viable.
Autonomous ground vehicles and robots deploying to new operational theaters produce accurate 3D occupancy maps from day one — without a retraining cycle that could take weeks and requires locally-collected labeled data that does not exist.
WHAT VIS-OCCANY DELIVERS
VIS-OCCANY (codename: Izanagi) implements OccAny (CVPR 2026, arXiv:2603.23502), a generalized unconstrained 3D occupancy prediction model designed explicitly for zero-shot generalization. Rather than learning city-specific or sensor-specific priors, OccAny learns scene-agnostic occupancy representations that transfer across environments without fine-tuning.
CAPABILITIES
- VIS-OCCANY (codename: Izanagi) implements OccAny (CVPR 2026, arXiv:2603.23502), a generalized unconstrained 3D occupancy prediction model designed explicitly for zero-shot generalization
- Rather than learning city-specific or sensor-specific priors, OccAny learns scene-agnostic occupancy representations that transfer across environments without fine-tuning
- CUDA, MLX, and CPU backends allow deployment from data center to edge without code changes.
WHY THIS IS HARD
Building VIS-OCCANY requires solving multiple coupled problems:
- 01Autonomous vehicles and robotic platforms deployed to new cities, new terrain types, or new operational theaters fail on distribution shift
- 023D occupancy models trained in European urban environments misread Middle Eastern cities; models trained on highways have no concept of unstructured intersections
- 03VIS-OCCANY (codename: Izanagi) implements OccAny (CVPR 2026, arXiv:2603.23502), a generalized unconstrained 3D occupancy prediction model designed explicitly for zero-shot generalization
- 04Rather than learning city-specific or sensor-specific priors, OccAny learns scene-agnostic occupancy representations that transfer across environments without fine-tuning
VIS-OCCANY solves these through careful architecture design and rigorous validation.
PROOF, NOT PROMISES
Key performance metrics:
| METRIC | VALUE |
|---|---|
| Zero-Shot Transfer | Generalizes to unseen environments without domain adaptation or retraining |
| Backend Coverage | CUDA + MLX + CPU — single model file runs from cloud server to Jetson Orin |
| Recognition | CVPR 2026 — top-tier peer review for a safety-critical perception task |
| Defense Angle | Day-one occupancy mapping in unfamiliar operational theaters; forward-deployed UGV and UAV perception without pre-mission data collection |
WHAT'S BUILT TODAY
| COMPONENT | STATUS | NOTES |
|---|---|---|
| Zero-Shot Transfer | COMPLETE | Generalizes to unseen environments without domain adaptation or retraining |
| Backend Coverage | COMPLETE | CUDA + MLX + CPU — single model file runs from cloud server to Jetson Orin |
| Recognition | IN PROGRESS | CVPR 2026 — top-tier peer review for a safety-critical perception task |
| Defense Angle | IN PROGRESS | Day-one occupancy mapping in unfamiliar operational theaters; forward-deployed UGV and UAV perception without pre-mission data collection |
WHERE VIS-OCCANY DEPLOYS
- APP_01
AUTONOMOUS SYSTEMS
Autonomous ground vehicles and robots deploying to new operational theaters produce accurate 3D occupancy maps from day one — without a retraining cycle that could take weeks and requires locally-collected labeled data that does not exist.
- APP_02
RESEARCH LABS
VIS-OCCANY (codename: Izanagi) implements OccAny (CVPR 2026, arXiv:2603.23502), a generalized unconstrained 3D occupancy prediction model designed explicitly for zero-shot generalization.
- APP_03
EDGE COMPUTING
CVPR 2026 occupancy model that predicts full 3D voxel occupancy in unseen environments — zero-shot, no domain adaptation required.