Skip to content

VGGSfM

Recover camera poses and 3D structure from 2D images

Visit official site

What VGGSfM is

The technique is described as a deep learning approach that recovers camera poses and 3D structure from a set of unconstrained 2D images through a fully differentiable, end-to-end framework.

What you can do with it

  • Extract 2D trajectories from input images
  • Recover cameras from image and trajectory features
  • Initialise a point cloud from trajectories and cameras
  • Refine reconstruction with a bundle adjustment layer

Example uses

  • Reconstruct scenes on the CO3D dataset
  • Recover cameras and point clouds on IMC Phototourism
  • Rebuild camera poses on the ETH3D dataset

Pros & cons

✓ What we like

  • The framework is fully differentiable end to end
  • Unconstrained photo collections are named as input

! What to watch out for

  • Reconstruction starts from a set of 2D images
  • The named evaluations use three public datasets

FAQ

Which datasets are named?

CO3D, IMC Phototourism and ETH3D are listed.

What framework is used?

A fully differentiable framework with a bundle adjustment layer.

Last reviewed: 2026-11-15

More AI image generation tools

View all →

How we review