SynFlow: Synthetic LiDAR Scene Flow Data for Autonomous Driving

A synthetic LiDAR scene flow benchmark collected using the CARLA simulator and formatted for compatibility with OpenSceneFlow. The dataset is provided by KTH and was last updated on June 20, 2026. It includes two pretrained model checkpoints to support further research.

Use Cases

• Training and benchmarking scene flow estimation models based on synthetic LiDAR data.
• Evaluating the transfer of models trained on synthetic data to real-world scenarios.
• Developing and testing algorithms for 3D motion understanding in autonomous driving.

Strengths

• Formatted for seamless compatibility with the OpenSceneFlow benchmark.
• Includes two pretrained model checkpoints (one trained on SynFlow-4k, another augmented with real-world data).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and file size are unknown, which may limit suitability assessment.

Provenance

Source

KTH, generated using the CARLA simulator.

Collection Method

Synthetic data collection via simulation.

Freshness

Last updated 2026-06-20 10:20:28; freshness should be verified.