iamdyeus/Vehicle-Scan-Preview-Meshes

Vehicle Scan Preview Meshes

A collection of 40 vehicle surface meshes in OBJ, STL and GLB, with a searchable vehicle catalog, geometry diagnostics and preview images. Prepared for the PrintYourVehicle community.

These are lightweight scan-derived web-preview meshes, intended as starting points for modelling, retopology, mesh-repair experiments and broad vehicle-shape reference. They are not the original high-resolution scans, verified engineering measurements, CAD solids or ready-to-print models.

Download and choose a format

Folder / file	Contents	Units / purpose
obj/	40 indexed triangle OBJ files	Source coordinates, interpreted as metres; preferred for further mesh work
stl/	40 binary STL files	Coordinates multiplied by 1,000; import as millimetres
glb/	40 byte-for-byte original GLBs with renamed files	Preserved source geometry and source coordinate system
car_list.csv	Exactly make,model,year, 40 rows	Simple vehicle catalog
metadata/manifest.csv	File paths, generation, year basis and identification notes	Connect catalog entries to downloadable models
metadata/mesh_quality.csv	Per-model geometry measurements and defects	Machine-readable audit
metadata/*.json	Equivalent detailed metadata and audit data	Programmatic use
previews/	Four galleries covering all 40 cars and three multi-view inspections	Rendered directly from supplied triangles
QUALITY_REPORT.md	Inspection findings and suitability assessment	Read before fabrication work
SOURCES.md	Identification references and unresolved provenance	Metadata documentation
tools/convert.py	Conversion, read-back verification and audit script	Reproduce mesh exports
SHA256SUMS.txt	SHA-256 checksums of the other package files	Integrity verification

The ewc- prefix was removed from every model filename. All other filename tokens were retained, including r2, r3, etc.; these are treated as source revision labels, not vehicle years. Each car has the same file stem across its three formats.

What “highest quality conversion” means here

The supplied archive contained GLBs only. Each contains one static indexed triangle mesh, with positions and triangle indices but no textures, UVs, vertex colours, materials or stored normals.

Every source vertex and triangle was preserved in OBJ, including original vertex ordering, triangle ordering and winding. OBJ positions use 17 significant digits and were parsed back and compared exactly against the decoded source coordinates. No decimation, subdivision, smoothing, remeshing, hole filling or automatic repair was applied.

STL contains the same triangles, scaled from source metre coordinates into millimetres. Face normals are calculated from the existing winding, not used to repair it. Binary STL uses 32-bit float coordinates; the largest measured coordinate rounding difference across all exports is 0.000245 mm or less. This is export rounding, not scan accuracy. All STL triangle records were read back and checked against the expected float32-scaled source coordinates.

OBJ uses flat shading by default. Smooth shading in a viewer may improve appearance without changing vertices, but cannot restore measured surface detail. Subdivision creates interpolated geometry, not recovered scan information. There is no MTL because the inputs contain no material data.

Scale and orientation

GLB convention uses metres, and the observed vehicle-size coordinates are consistent with metres. Nevertheless, there are no calibration records or independently measured landmarks in this archive. 1:1 dimensional accuracy has not been validated.

• GLB and OBJ retain original positions and axes; the models appear to use X along vehicle length, Y vertical and Z across width.
• STL retains those axes and the original origin, with a uniform ×1,000 coordinate conversion. STL itself has no enforceable unit declaration: choose millimetres in the importing application.
• Do not multiply the STL coordinates by 1,000 again.
• Some applications expect Z-up. Set the import axis conversion or rotate the model appropriately; no rotation is baked into these exports.
• Check dimensions against known physical measurements before scaling a design. Bounding-box width may include mirrors, and open/missing geometry can affect measured extents.

A car about 4.5 units long in OBJ should be about 4,500 units long in the supplied STL. Both represent the same geometry at the intended physical scale. For a 1:24 model, scale the millimetre STL by 1/24 after checking its dimensions.

Vehicle names and years

The catalog identifies vehicles from their filenames and generation/chassis codes. year is deliberately a text field containing a generation range or starting year, not a claim about the exact car scanned.

• Most rows use US model-year ranges; a few use global production/generation ranges. The basis and market differences are explicit in metadata/manifest.csv.
• 2023+ means the identified generation starts with MY2023 and no terminal year is assigned here. It does not identify the scan year or promise compatibility with every later facelift.
• Exact scan years are unknown for all 40 models. Trims, market specifications, aftermarket modifications and facelift details are not established unless the filename supports them.
• The metadata is inferred rather than manufacturer-certified. A generation range does not imply that one mesh matches every model year in that range.
• Known conflict: toyota-gr86-zd8-r4 says Toyota GR86, but Toyota GR86 uses ZN8 while ZD8 is Subaru BRZ. Its catalog entry remains Toyota/GR86 provisionally, with the conflict stated. The filename is retained for traceability; independent source confirmation is needed.

Quality at a glance

The dataset has 1,187,961 triangles in total, with 25,580–30,000 triangles per model. All 40 models contain boundary edges and non-manifold edges. None passes the audit's closed edge-manifold check. Wheels, glass, openings and underbody coverage vary between models; some have only partial wheel geometry or empty arches.

These meshes can be useful for:

• Understanding broad body proportions and building visual concepts.
• Starting a manually checked retopology or surface reconstruction.
• Developing algorithms for scan cleanup, hole detection and mesh repair.
• Creating display models after additional modelling, thickness and print validation.

They should not be treated as validated mounting-point data, tooling surfaces, CFD-ready boundaries or direct full-size panel-production geometry. Validate local dimensions against the actual vehicle or a calibrated scan before using them for fitment.

See QUALITY_REPORT.md for measured statistics, methods, limitations and per-car results.

Gallery

The renders show actual supplied geometry with flat-shaded surfaces. They are visual inspection aids, not enhanced reconstructions. Triangle sorting and overlapping surfaces can produce extra visual clutter; measured topology defects are reported separately.

Vehicles 11–20 · Vehicles 21–30 · Vehicles 31–40