HELVIPAD: A Real World Dataset for Omnidirectional Stereo Depth Estimation The Helvipad dataset is a real world stereo dataset designed for omnidirectional depth estimation. It comprises 39,553 paired equirectangular images captured using a top bottom 360° camera setup and corresponding pixel wise depth and disparity labels derived from LiDAR point clouds. The dataset spans diverse indoor and outdoor scenes under varying lighting conditions, including night time environments. News [⚠️ Important Update – 21/09/2025] A minor error was identified in the Helvipad depth to disparity conversion formula. We have now regenerated and uploaded the correct disparity maps to the repo. The previous disparity maps are still included, since they were used in the experiments reported in the Helvipad paper and the DFI OmniStereo paper. For new work, we recommend using the corrected disparity maps. [16/02/2025] Helvipad has been accepted to CVPR 2025! 🎉🎉 Dataset Structure The dataset is organized into training, validation and testing subsets with the following structure: The dataset repository also includes: helvipad utils.py : utility functions for reading depth and disparity maps, converting dis…
We use cookies for essential functionality and analytics. You can accept or reject analytics cookies.Cookie policy