m monad-knowledge FIIT STU · open research vault
dataset

fiit-library-roomplan-2026-05-29

Also known as: FIIT Library RoomPlan scan, fiit-library-roomplan-2026-05-29, library-lidar

Description

First-party indoor 3D capture of the FIIT STU library reading hall (the main room), taken 2026-05-29 with Apple RoomPlan (via the Polycam app) on an iPhone/iPad Pro LiDAR sensor. Unlike the public CSI/pedestrian datasets in this folder, this is own-captured geometry — the source of the registered floor fiit-library-floor-0 and the furniture/seat layer added under IP-109 (furniture-rich LiDAR 3D).

The capture is a multi-format bundle, not a single file: a dense colored point cloud, a parametric RoomPlan model (a mesh of 273 named objects — walls, doors, windows, and 103 semantically-classified furniture instances), a CAD floor plan, and 2D renders. The parametric layer is what makes this dataset valuable: Apple RoomPlan did on-device instance segmentation + classification, so each chair/table/shelf arrives labelled with an oriented bounding box and height.

Classification

  • Kind: indoor 3D LiDAR scan + parametric room model (Apple RoomPlan).
  • Modality: dense RGB point cloud (~391 k–1.1 M points, ~20–62 mm spacing) + semantic mesh + 2D CAD.
  • Scene: single room, ~27.3 × 18.1 m bounding box, 338.6 m² livable, 4.0 m ceiling, 1349 m³, 79 m perimeter, 4 doors, windows.
  • RoomPlan inventory (from the CSV): 103 furniture instances — 68 chairs (37 dining / 17 dining-with-arms / 14 other), 21 tables (12 dining-rect + 5 coffee + 3 other-rect + 1 dining-elliptic), 7 shelves, 3 cabinets, 3 sofas, 1 TV.
  • Frame: scan-local metres (PLY Z-up; OBJ Y-up, ARKit). Registered into SRID 900010 (local Cartesian cm) by the transform in cache_to_900010.json (NB 07).

Files (manifest)

Stored at s3://monad-knowledge/datasets/fiit-library-roomplan-2026-05-29/ (filenames preserved — the OBJ references 29_05_2026.mtl, the MTL references textures/floor_Living_Room_color.jpg, so renaming individual members would break the bundle). SHA-256 (first 12) + role:

File Bytes SHA-256 Role
29_05_2026.ply 10,560,180 2aa70c99b1d3 Dense colored point cloud (binary PLY, XYZ+RGB)
29_05_2026.pts 22,675,835 ec114ad07c30 Dense point cloud (ASCII PTS, XYZ+intensity+RGB)
29_05_2026.laz 3,331,103 3684ba0e44ba Point cloud (LASzip-compressed LAS)
29_05_2026.obj 5,111,905 e07a5770c90f Parametric RoomPlan mesh — 273 named objects (120 walls, 8 doors, 2 windows, 2 floor, 103 furniture)
29_05_2026.mtl 8,601 08d3f27dfc03 OBJ material library
29_05_2026.csv 8,376 14da83ef51b0 RoomPlan inventory — per-instance oriented dims + room totals + GPS
29_05_2026.dxf 3,302,751 747857714dad CAD floor plan — layered (Poly-Walls/Doors/Windows/Furniture, 10k+ polylines)
29_05_2026.svg 2,245,307 b01af91d99fd 2D floor plan (vector)
29_05_2026.pdf 2,308,342 37bc16d4b598 2D floor plan (print)
29_05_2026.png 845,514 120a083a686f 2D floor plan (raster preview)
textures/floor_Living_Room_color.jpg 97,213 c56fe28f2c19 Floor texture (referenced by the MTL)

manifest.json (full SHA-256s + sizes) is uploaded alongside the bundle. Local working copy: _cache/library-lidar/ (gitignored; S3 is the archival home).

Used by

  • IP-109 (furniture-rich LiDAR 3D) — the furniture-rich 3D pipeline; the OBJ/CSV/DXF parametric layer is parsed in notebooks/floorplans/lidar/06_parse_parametric.ipynb, registered in 07_align_bundle.ipynb, reduced in 08_furniture_methodologies.ipynb, and ingested into PostGIS in 09_furniture_to_gis.ipynb (103 furniture obstacles + 68 chair-seats on fiit-library-floor-0).
  • The registered point cloud derivatives live beside the GIS source: _attachments/floorplans/fiit-library/ (29_05_2026.clean.ply, register.json, furniture.geojson, …).
  • Floor fiit-library-floor-0 in the PostGIS indoor schema (SRID 900010).

Notes

  • Quality gate (IP-109 Q3): OBJ↔CSV recall 100 % (103=103), registration residual ~9 cm, point spacing ~20 mm — clears the per-room bar for accepting a scan. See quality_gate.json beside the GIS source.
  • The same physical scan was exported twice by Polycam: this bundle (391 k-point cloud + full parametric/CAD set) and the floorplan-dir copy (1.1 M-point cloud, no parametric set). The OBJ here shares the floorplan copy's exact frame (matched to 6 mm), which is why furniture reuses the tracked transform.
  • Related diary: 2026-05-31 - iPhone LiDAR scan quality and the 3D question, 2026-06-26 - RoomPlan furniture beats DBSCAN blobs.
  • More rooms: drop a new Polycam export, re-run NB 06–09, and add a sibling dataset note fiit-<room>-roomplan-<date> under the same S3 datasets/ prefix.