The first open-source collaborative perception dataset focused on adverse weather conditions.
Key Features
- Annotation schema compatible with OPV2V.
- 110 scenarios including:
- 5 road configurations.
- 11 weather/daytime conditions.
- 2 density settings.
- Every scenario features collaboration between 5 viewpoints:
- The main vehicle (Ego).
- 2 Roadside Units (RSU 1 and RSU 2).
- 2 Connected Vehicles (Car 1 and Car 2).
Road Configurations
Each configuration was chosen from real-world data accounting for challenging conditions involving adverse weather and poor visibility.
Weather Conditions
Having a balanced distribution of diverse conditions, Adver-City is the first synthetic collaborative perception dataset to introduce glare.
Density Settings
The varying number of objects in each setting enables novel experimentation with different levels of occlusion.
Sensor Suite
Balancing performance and real-world feasibility, our sensor suite is based on configurations commonly used in practice and enables Sim2Real approaches.
| Sensors | Details |
|---|---|
| 4x RGB cameras | Resolution: 1920 x 1080 pixels, HFOV: 100 degrees |
| 4x Semantic cameras | Resolution: 1920 x 1080 pixels, HFOV: 100 degrees |
| 1x 3D LiDAR | Channels: 32 channels, Points Per Second: 1.2M, Frequency: 10 Hz, Range: 200 meters, VFOV: -25 to 15 degrees |
| GNSS & IMU | Standard deviation: 3e-6 rad (latitude and longitude) and 0.05 meters |
