Skip to main content
GIS-Schools GIS-Schools
3D Reconstruction ยท GeoAI

MASt3R - Image Matching and 3D Reconstruction

Explore learned image matching and point-map reconstruction from overlapping photos, with explicit scale, georeferencing and licensing checks.

3D Reconstruction task illustration
Family: Learned multi-view geometryArchitecture: ViT encoder with point-map and matching headsVersion: ViTLarge BaseDecoder 512 metricLicense: CC BY-NC-SA 4.0 plus checkpoint dataset restrictions
MASt3R - Image Matching and 3D Reconstruction

Exact implementation and checkpoint

Selected official model: MASt3R_ViTLarge_BaseDecoder_512_catmlpdpt_metric. This page follows MASt3R and its matching/global-alignment workflow, not an interchangeable DUNE variant.

How it works

The network predicts geometry and descriptors from image pairs. Matches and global alignment connect observations into a reconstructed scene.

Inputs and outputs

Overlapping RGB views enter the documented image loader. Outputs support 3D point maps, image correspondences and scene alignment. A usable textured mesh or BIM model requires additional processing.

Training and fine-tuning

Follow the repository training documentation using datasets for which you have rights. Preserve the image-processing settings and checkpoint configuration. Begin by checking a small multi-view scene against independent measurements.

GIS and digital twin use

A research starting point for visual documentation and reconstruction experiments around infrastructure. Establish scale and georeferencing, inspect missing geometry and then prepare an appropriate reality-data format.

Validation on your data

Compare reconstructed distances and control points against independent survey evidence. Review overlap coverage, camera alignment and outlier points before using measurements.

Limitations

Textureless, reflective and changing surfaces can fail. A metric checkpoint name is not proof of survey accuracy or an absolute CRS. Large scenes require careful image pairing and compute planning.

Resources and license

The official code is non-commercial CC BY-NC-SA 4.0. Checkpoint use also depends on training-dataset conditions, including restrictive notices identified by the authors. Official links are provided without accepting terms or redistributing binaries.

Integration and validation status

This is a workflow component, not an installed iTwin plugin or automatic asset update service. Preserve coordinate reference systems, units, acquisition dates and source provenance for spatial outputs; preserve timestamps and sensor units for time-series outputs. Review results before attaching them to an authoritative asset.

GISSchools has not run training, inference or an accuracy benchmark for this entry. Covers and diagrams are conceptual illustrations. Official resources remain external.

Official references

Examples and images

Downloads and resources

External resourceOfficial repositoryProvider: NAVER Labs
Open resource ↗
External resourceUsage and configurationProvider: NAVER Labs
Open resource ↗
External resourceOfficial Checkpoints and NoticesProvider: NAVER LabsVersion: ViTLarge BaseDecoder 512 metricLicense: CC BY-NC-SA 4.0 plus checkpoint dataset restrictions

The official code is non-commercial CC BY-NC-SA 4.0. Checkpoint use also depends on training-dataset conditions, including restrictive notices identified by the authors. Official links are provided without accepting terms or redistributing binaries.

Open resource ↗
External resourceLicense and noticesProvider: NAVER Labs
Open resource ↗
Discussion

Comments

0

No comments yet. Start the discussion.

Join the discussion

Leave a reply

Your email address will not be published. Required fields are marked.