Skip to content

DL3DV-10K

A large real-scene video dataset for 3D vision

Visit official site

What DL3DV-10K is

The dataset is described as a large collection of real-scene videos for deep learning research in three-dimensional vision, with manual annotation on every clip.

What you can do with it

  • Optimise NeRF model performance
  • Train vision-language models
  • Develop SLAM systems from viewpoint and scene information

Dataset content

  • More than 10000 high-quality videos
  • Manual labels for key points and environment complexity
  • Camera poses, NeRF depth, point clouds and 3D meshes

Pros & cons

✓ What we like

  • Supports NeRF, tracking and vision-language research
  • Scene viewpoints feed SLAM system development

! What to watch out for

  • Annotations are done manually per video
  • Depth is described as NeRF estimated

FAQ

How many videos are included?

The material states more than 10000 high-quality videos.

What data comes with each video?

Camera poses, NeRF estimated depth, point clouds and 3D meshes are listed.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review