Wheelchair accessibility audit from a 3D scan

Data for Wheelchair accessibility audit from a 3D scan.

The code lives in wheelchair-accessibility-audit/. The full write-up and the one-click downloads are at https://learngeodata.eu/materials/wheelchair-accessibility-audit/.

• issues.csv The 26 numbered issues, with room, position, measured and required value.
• metrics.json The headline numbers in one flat file, with the SHA-256 fingerprint of the scan the audit read.
• summary.json Every number the audit printed, with the door list and the issue list.

Wall thickness from a scan, written to IFC

Data for Wall thickness from a scan, written to IFC.

The code lives in wall-thickness-to-ifc/. The full write-up and the one-click downloads are at https://learngeodata.eu/materials/wall-thickness-to-ifc/.

• measured_partitions.ifc The three walls as IFC4 walls with their measured thickness, 166 entities, schema-valid. 11 kB.
• result.json Every face, pair, thickness and check the article quotes, including the density profiles.

Scan to floor plan, from wall pixels to a DXF

Data for Scan to floor plan, from wall pixels to a DXF.

The code lives in scan-to-floor-plan-dxf/. The full write-up and the one-click downloads are at https://learngeodata.eu/materials/scan-to-floor-plan-dxf/.

• classroom_scene.ply The classified classroom, 1,422,440 points with colour and a scalar_semantic_label field, 27 MB.
• floor_plan.dxf The drawing in metres, six layers, evidence on every wall, the file your run should reproduce. 55 kB.
• metrics.json Every number the article quotes, step by step.

Open-vocabulary 3D semantics on a real room scan

Data for Open-vocabulary 3D semantics on a real room scan.

The code lives in open-vocabulary-3d-semantics/. The full write-up and the one-click downloads are at https://learngeodata.eu/materials/open-vocabulary-3d-semantics/.

• room.ply The living-room scan, thinned to 999,987 points (one per 1.3 cm voxel) with colour and a class on each, 19 MB.
• room_labeled.ply The 40,000 working points, levelled and in metres, with the pipeline's label and the reference class on each.
• floorplan_slice.ply The 30 cm slab at 1 m, the plan slice your run should reproduce.
• class_stats.csv Per-class points, heights, footprint along the walls and agreement with the reference.

Label a 3D scene by painting on photos

Data for Label a 3D scene by painting on photos.

The code lives in 3d-scene-labeling-gui/. The full write-up and the one-click downloads are at https://learngeodata.eu/materials/3d-scene-labeling-gui/.

• reconstruction_data.npz The KIDS reconstruction from the 3d-models-from-images kit, 9 frames and 1,270,080 points. 44 MB.
• smart_fused_labels-v2.ply All 1,270,080 points with their colour and a segment_label field CloudCompare reads as a scalar field. 24 MB.
• semantic_scene.glb The labeled scene with the nine camera frustums, for any glTF viewer. 20 MB.

3D models from photos with Depth-Anything-3

Data for 3D models from photos with Depth-Anything-3.

The code lives in 3d-models-from-images/. The full write-up and the one-click downloads are at https://learngeodata.eu/materials/3d-models-from-images/.

• kids_photos.zip Nine phone photos of a toy construction set, 4000 x 2252, EXIF removed, 21 MB.
• reconstruction_data.npz Depth, confidence, cameras, images and 1,270,080 points. The scene-labeling kit starts here. 44 MB.
• kids_reconstruction_preview.glb The cleaned point cloud, downsampled to 250,000 points, to open in any glTF viewer. 4 MB.

LiDAR ground filtering to a terrain model

Data for LiDAR ground filtering to a terrain model.

The code lives in lidar-ground-to-dtm/. The full write-up and the one-click downloads are at https://learngeodata.eu/materials/lidar-ground-to-dtm/.

• lidar_hd_tile.laz 300 x 300 m of IGN LiDAR HD, 3,830,133 points with IGN's classes, 21 MB.
• dtm.tif The 0.5 m terrain model (GeoTIFF, EPSG:2154) the script should reproduce.
• ground.laz The 1,409,542 points the filter called ground, IGN's class kept on each.

Gaussian splat to textured mesh

Data for Gaussian splat to textured mesh.

The code lives in gaussian-splat-to-mesh/. The full write-up and the one-click downloads are at https://learngeodata.eu/materials/gaussian-splat-to-mesh/.

• cabin_in_the_forest.ply The trained splat, 84,528 Gaussians, 20 MB.
• cabin_textured.glb The finished mesh, so you can check your result against mine.