Datasetpublic

OSResight/tartanair-episodes-v3-parquet

tartanair-episodes-v3-parquet Per-frame Parquet dataset for ReCAST tracker training. Schema One row per frame, grouped by episode_id. Arrow memory-mapped access enables reading specific frames without loading entire episodes. Column Type Description episode_id int32 Episode identifier frame_idx int32 Frame index within episode jpeg binary JPEG-encoded RGB frame depth list<float32> Flat H×W depth map seg list<uint16> Semantic segmentation (empty… See the full description on the dataset page: https://huggingface.co/datasets/OSResight/tartanair-episodes-v3-parquet.

sourceHugging Facebsd-3-clauseupdated 6mo agoView on Hugging Face

0likes443downloads

Dataset Card

tartanair-episodes-v3-parquet

Per-frame Parquet dataset for ReCAST tracker training.

Schema

One row per frame, grouped by episode_id. Arrow memory-mapped access enables reading specific frames without loading entire episodes.

Column	Type	Description
episode_id	int32	Episode identifier
frame_idx	int32	Frame index within episode
jpeg	binary	JPEG-encoded RGB frame
depth	list\<float32\>	Flat H×W depth map
seg	list\<uint16\>	Semantic segmentation (empty if absent)
pose	fixed\size\list\<float64, 16\>	4×4 camera-to-world pose (vision convention)
intrinsics	fixed\size\list\<float32, 4\>	[fx, fy, cx, cy] in pixels
gravity	fixed\size\list\<float32, 3\>	World-frame gravity direction
camera_model	string	"pinhole" or "equidistant"

Conventions

—Poses: Vision convention (X=right, Y=down, Z=forward). Normalized during conversion.
—Depth: Projective z-depth (all datasets, normalized during conversion).
—Gravity: Unit vector pointing downward in world frame.

Provenance

—Dataset: tartanair
—Episodes: 60
—Frames: 72000
—Created: 2026-03-15T14:19:10.370555+00:00
—Converter: scripts/convert_episodes_to_parquet.py