ClearDepth: Synthetic Transparent-Object Stereo Video and Labels

ClearDepth Transparent (FiftyOne) is a grouped dataset built from the synthetic transparent-object stereo data described in the ClearDepth paper. It contains indoor scenes with stereo RGB video, dense per-pixel labels, and optional merged 3D point-cloud reconstructions. The dataset was created by Voxel51.

Use Cases

• Train depth estimation models based on synthetic stereo RGB video data.
• Evaluate 3D reconstruction algorithms based on the optional merged point-cloud data.
• Develop models for transparent object segmentation based on dense per-pixel labels.
• Benchmark stereo vision systems using synchronized indoor scene video sequences.

Strengths

• Dataset is grouped by indoor scene, providing a structured view of each environment.
• Includes multiple synchronized data modalities: stereo RGB video, dense per-pixel labels, and optional 3D point clouds.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Voxel51

Collection Method

Synthetic data generation as described in the ClearDepth (SynClearDepth) paper.

Freshness

Last updated 2026-05-26 18:53:24; freshness should be verified.