PanoHK360 is a city-scale outdoor panoramic dataset captured across diverse urban environments in Hong Kong. It combines 8K equirectangular RGB panoramas with LiDAR-derived metric depth, surface normal maps, raw point clouds, and 6-DoF poses. The dataset was created by adadai3132 and a filtered sample release is available on Hugging Face.
Use Cases
- Train depth estimation models based on the paired 8K RGB panoramas and LiDAR-derived metric depth.
- Develop algorithms for 3D scene reconstruction based on the provided raw point clouds and 6-DoF poses.
- Benchmark surface normal prediction models using the included surface normal maps.
- Simulate urban navigation and planning based on continuous outdoor capture sequences across Hong Kong.
- Evaluate panoramic image processing techniques on high-resolution 8K equirectangular images.
Strengths
- Includes high-resolution 8K equirectangular RGB panoramas.
- Provides LiDAR-derived metric depth data, offering accurate ground truth.
- Contains multiple complementary data types: surface normal maps, raw point clouds, and 6-DoF poses.
- Captured across diverse urban environments in Hong Kong, suggesting geographic variety.
Limitations
- Description metadata is limited; actual data quality requires manual inspection after download.
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown, which may limit suitability assessment.
Provenance
- Source
- adadai3132 on Hugging Face.
- Collection Method
- Captured across diverse urban environments in Hong Kong, likely using panoramic cameras and LiDAR sensors.
- Freshness
- Last updated 2026-05-07 11:03:52; freshness should be verified.
- Geography
- Hong Kong.