Cv Ds All: Autonomous Driving Novel-View Synthesis Dataset

MHDCSM/cv_ds_all is a WebDataset-format collection of shards for autonomous driving novel-view-synthesis. The dataset includes training samples with ground-truth target images and separate test samples without targets. Each sample contains camera poses, intrinsics, timestamps, six camera views per timestamp, and dense world-frame point clouds.

Use Cases

• Train novel-view-synthesis models based on multi-camera images and camera pose data.
• Evaluate 3D scene reconstruction algorithms based on dense LiDAR point clouds.
• Develop multi-sensor fusion models for autonomous driving using synchronized camera and LiDAR data.
• Benchmark camera pose estimation techniques using provided intrinsics and timestamps.

Strengths

• Data is organized in WebDataset-format shards, which is a standard for efficient large-scale ML training.
• Samples include synchronized multi-modal data: six camera views per timestamp, camera poses, intrinsics, timestamps, and LiDAR point clouds.
• Dataset provides a clear separation between training samples (with ground-truth targets) and test samples (without targets).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Last updated 2026-05-22 23:34:22; freshness should be verified.

Provenance

Source

MHDCSM

Freshness

Last updated 2026-05-22 23:34:22.