Download¶
The full RealSim-CP dataset is hosted on Google Drive.
:material-google-drive: Download RealSim-CP
License
The dataset is released under CC BY 4.0 — free to use, including commercially, with attribution. Please cite the paper (see README → Citation). The tools in this repository are MIT-licensed.
What you download¶
The dataset is organized by scenario, then by weather/time variant:
RealSim-CP/
└── <scenario>/ # e.g. daiba_station_scenario
├── sunny_1/ sunny_2/ # clear daytime
├── night_1/ night_2/ # clear nighttime
└── rainy_1/ rainy_2/ # rainy daytime
Each variant is a self-contained ~10-second clip at 10 Hz with its own images, point clouds, and a single OpenLABEL annotation file. See Dataset structure for the per-variant layout.
Grab only what you need¶
Variants are independent, so you can download a single weather/time variant of a single scenario to get started (a few GB) rather than the whole dataset. A good first download is one night variant of the Daiba Station scenario — it has the most complete sensor coverage.
Recommended first scene
daiba_station_scenario/night_1 — 4 RSUs + 11 vehicles, full camera and
LiDAR coverage, 101 frames. Pair it with the
Quickstart and the
Visualizer.
Approximate sizes¶
A single fully-populated variant (images for all agents + LiDAR for all agents + the label file) is on the order of a few gigabytes; image-heavy multi-vehicle variants are larger. Daytime variants may omit some LiDAR (see Known issues). Plan disk accordingly before pulling multiple scenarios.
Next steps¶
- Dataset structure — folders, agents, and sensor streams.
- Sensors — camera and LiDAR specifications.
- Label format — OpenLABEL schema and coordinate frames.
- Loading the data — Python loading code.