(5/5) Explore Helvipad:
🌐 Project page: vita-epfl.github.io/Helvipad
📚 Paper: arxiv.org/abs/2411.18335
🤗 Dataset: huggingface.co/datasets/chc...

(4/5) We benchmarked leading models and introduced crucial adaptations tailored to omnidirectional imagery—significantly enhancing accuracy and robustness.

(3/5) Helvipad includes:
🔹40K paired top-bottom omnidirectional images
🔹 29 diverse indoor/outdoor scenes (busy corridors, squares, footpaths)
🔹 High-quality depth and disparity annotations

(2/5) Accurate depth perception is vital for robotic navigation, but existing methods often rely on expensive LiDAR sensors or limited-view cameras.