DL3DV-10K: 10,000 Multi-View Scenes with 480P Frames and Camera Poses

DL3DV-10K subset containing 480P resolution video frames and corresponding camera poses for 3D vision tasks. Released by the DL3DV team in 2024, it provides a large-scale collection of multi-view imagery for novel view synthesis. The data is specifically formatted to support efficient training of neural rendering models.

Use Cases

• Novel view synthesis using the 480P frames and camera poses
• Training 3D Gaussian Splatting models on the multi-view image sequences
• Evaluating camera pose estimation algorithms against the provided pose ground truth

Strengths

• Includes 10,000 distinct scenes from the DL3DV-10K collection
• Provides pre-computed camera poses for every frame
• Standardized 480P resolution for reduced computational overhead

Limitations

• Resolution is capped at 480P, which lacks the fine detail of the 960P or original versions
• Ongoing review process for sensitive information may result in filtered or missing content

Provenance

Source

DL3DV

Freshness

Last updated September 2024.