Open-vocabulary panoramic segmentation across diverse embodied platforms.
🔭 Overview · 🎬 Demo · 🗂️ Benchmark · 🚀 Release Status
EmbPASS introduces Cross-Embodiment Open Panoramic Segmentation for open-vocabulary panoramic perception across heterogeneous embodied platforms.
It establishes a multi-platform panoramic semantic segmentation benchmark spanning Vehicle, Drone, Wearable, and Quadruped, and proposes EPONet for panoramic perception under heterogeneous embodied observations.
Overview of EmbPASS and cross-embodiment open panoramic segmentation.
embpass_readme.mp4
Panoramic observations from diverse embodied platforms and the corresponding segmentation results.
EmbPASS consists of 1,000 annotated panoramic images spanning four representative embodied platforms: Vehicle, Drone, Wearable, and Quadruped, under a unified semantic taxonomy.
🚧 The EmbPASS benchmark and EPONet source code will be made publicly available.
- 📚 Release the EmbPASS benchmark.
- 💻 Release the EPONet source code.