3D Tiles translation is the localization of public-facing names, descriptions, legends and metadata in a streamed 3D geospatial scene without moving buildings, altering feature identities or damaging tile loading. OGC 3D Tiles 1.1 stores a hierarchy of tiles with bounding volumes, selection properties, content references and often glTF assets. A city can be understood in several languages while remaining the same measurable, selectable three-dimensional environment. That outcome requires translators to distinguish language from geometry, metadata schema and rendering instructions.
If you need to localize 3D Tiles 1.1, translate 3D geospatial building labels, preserve glTF feature metadata, understand boundingVolume and geometricError, work with implicit tiling, or translate 3D Tiles style legends safely, this guide provides an end-to-end method. The central rule is to preserve structural keys, numerical properties, feature IDs, transformations, URI references and controlled values. Translate only the content that an authorized data owner has identified as human language, and validate it in the actual viewer.
This article is designed for geospatial engineers, GIS teams, digital-twin publishers, visualization developers, museums, city governments, localization professionals and educators worldwide. It concentrates on 3D Tiles streaming and glTF-linked metadata, not the CityJSON or CityGML authoring models and not ordinary 2D vector tiles. We will move from files and tile trees to feature selection, multilingual display, field-level quality control and reliable releases, using practical failure examples to show why each distinction matters.
3D Tiles is a streaming structure for large three-dimensional scenes
OGC 3D Tiles organizes geospatial 3D content into a spatial hierarchy so clients can fetch appropriate portions as needed. It is not a collection of paragraphs awaiting translation.
Worked example: A digital twin serves buildings and terrain to a phone. The viewer requests lower-detail tiles first, then more detailed tiles near the user’s viewpoint; language labels are a separate concern.
Failure to avoid: A translator who edits tile-selection or content references to improve visible words can interrupt streaming even when a small static preview looks correct.
Validation: Verify the same scene extent and tile hierarchy load in the target application while approved captions and attribute labels appear in the requested language.
Record the specification version and asset profile
The OGC 3D Tiles 1.1 standard uses glTF as the primary tile-content format and retains compatibility provisions for legacy formats from earlier implementations.
Worked example: A project includes one new 1.1 tileset and a historical 1.0 archive. Their content formats and extension requirements differ, even though both appear under a directory named tiles.
Failure to avoid: Assuming that a tool understands every version or silently rewriting an asset version string can create a delivery that fails for some clients.
Validation: Record the tileset asset version, extensions and intended viewer versions before editing language fields. Validate against the actual delivery specification.
Recognize tileset.json as machine configuration
A tileset JSON document describes tileset metadata and a root tile with spatial and selection properties. Keys such as root, geometricError and boundingVolume are interface syntax.
Worked example: A developer translates a page’s title into Spanish but must leave the tileset root object and its recognized member names intact for the 3D viewer.
Failure to avoid: Renaming root to an ordinary-language equivalent or changing numeric properties under the impression that they are decorative settings stops software from interpreting the scene.
Validation: Parse the JSON as an actual tileset and compare all structural member names. Only approved human-readable metadata should differ across language editions.
Preserve the root tile and its descendants
Each tile can contain child references forming a spatial tree. That tree controls progressive loading and is more than an editorial table of contents.
Worked example: A city center has a coarse parent tile and fine-grained children for streets and buildings. A multilingual interface needs the same spatial decomposition.
Failure to avoid: Removing a child or rewriting a link can leave holes in coverage. A viewer screenshot captured from one camera angle may not reveal all missing branches.
Validation: Traverse the tree before and after localization, compare node counts and test traversal from several camera positions and zoom ranges.
Understand geometricError as a selection measure
The geometricError value helps clients decide when a tile is sufficiently detailed. It is a numeric rendering contract, not an adjective describing content quality.
Worked example: An overview tile has a larger error threshold than its detailed descendants. A target-language page may explain this tradeoff without changing the thresholds.
Failure to avoid: Replacing the field with translated text or adjusting its numbers to make the map seem sharper can produce excessive requests or unacceptably coarse visualization.
Validation: Check the numeric values across the hierarchy and verify that the client selects a sensible set of tiles at the same camera positions.
Respect refine behavior ADD and REPLACE
Refinement declares whether a tile’s descendants add to its presentation or replace its detail. This affects which content stays visible while navigating.
Worked example: A tileset uses REPLACE for progressively finer city surfaces. An annotation translator must not switch it to ADD merely because they want to show more text.
Failure to avoid: The wrong refinement mode can display duplicate overlapping geometry, disappear features, or create a scene that looks inconsistent during zoom transitions.
Validation: Compare refine settings at their actual places in the hierarchy and observe scene transitions through a range of viewpoints.
Keep bounding volumes immutable
Bounding volumes contain spatial extents such as box, sphere or region. They help a viewer decide what to load and where content belongs.
Worked example: A high-rise building is inside a bounding region. A French caption for the building does not justify changing its box dimensions, latitude or elevation ranges.
Failure to avoid: Minor changes to bounding values can lead to aggressive culling, missing objects or excessive streaming. A valid JSON file is not proof of correct spatial containment.
Validation: Compare bounding-volume arrays exactly and confirm that the original tiles remain within their declared extents.
Distinguish box sphere and region encodings
3D Tiles can represent bounding volumes in different forms. A box is oriented in three-dimensional space, a sphere has center and radius, and a region encodes geographic intervals.
Worked example: An engineer recognizes a region from west, south, east, north and height bounds. The values are coordinates and measurements, not comma-separated prose to localize.
Failure to avoid: Treating region entries as degrees when their specification uses radians can place entire tilesets incorrectly. Reordering components can have a similarly severe effect.
Validation: Use the relevant standard for each volume type and test bounding extents against known geographic control points before and after export.
Preserve tile transforms and matrix order
Tile transforms are numeric matrices that position tile content relative to its spatial context. Translation of place labels must not alter transformation mathematics.
Worked example: A model positioned accurately above a road uses a 4-by-4 transform. A new-language facade sign can change as a texture asset, but the placement transform remains the same.
Failure to avoid: Rearranging matrix elements or changing translation components can rotate or displace structures. A small misalignment might not show until combined with survey data.
Validation: Compare all transform arrays and inspect overlap with reference geometry at more than one point in the scene.
Understand coordinate frames at format boundaries
glTF content uses its own coordinate conventions, and 3D Tiles applies transforms to place tile content in the geospatial world. The boundary requires careful software handling.
Worked example: A model exports correctly from a modeling tool but appears rotated after packaging. The source file may be fine; the coordinate-frame conversion may be wrong.
Failure to avoid: Trying to repair a frame mismatch by changing a translated location string cannot help. Arbitrary rotations made during language work hide a technical defect.
Validation: Test a recognizable vertical and horizontal reference after packaging. Keep coordinate engineering decisions separate from translation review.
Treat content.uri as a resource address
The uri property points to tile content or nested assets. It is an operational reference, not natural language merely because it contains readable words.
Worked example: A tileset references a GLB model stored under a path named buildings/north-block.glb. A tourist interface can translate North Block without renaming that resource path.
Failure to avoid: A file path translated without corresponding packaging and links creates broken requests. Case sensitivity and escaped characters add further deployment risks.
Validation: Keep URIs stable unless a deliberately planned asset migration updates and verifies every reference and server response.
Understand the contents array for multi-content tiles
A tile may refer to more than one content asset in supported 3D Tiles profiles. That allows different renderable layers to occupy a spatial tile.
Worked example: A district tile combines a built environment with another visual layer. The translator sees several content references but must not treat them as competing language variants.
Failure to avoid: Dropping one reference because it appears redundant removes real scene data. Confusing content multiplicity with linguistic alternatives damages the model.
Validation: Compare each referenced content and expected asset count, then test all layers in the target client.
Use glTF 2.0 as the main tile payload
glTF carries geometry, meshes, materials, textures and optional metadata for tile content. The file may be binary GLB or JSON with associated resources.
Worked example: A museum campus publishes a 3D model whose mesh remains unchanged while visitor-facing metadata and an approved sign texture vary by language.
Failure to avoid: Editing glTF buffer data or mesh accessors to alter text can unintentionally move vertices or corrupt binary alignment. The model is not an editable sentence document.
Validation: Preserve geometric buffers, accessors, node transforms and material assignments unless a qualified 3D asset engineer authorizes a separate asset adaptation.
Distinguish model text from texture images
Some visible writing in a 3D scene is part of an image texture, while other labels are generated from attribute metadata by the viewer.
Worked example: A building sign baked into a texture needs a localized image asset, whereas a popup title drawn from a metadata field needs a translated string.
Failure to avoid: A text localization plan that only searches JSON misses baked-in signage. Replacing a texture without mapping its UV coordinates may spoil the facade.
Validation: Inventory visible text by source: metadata, UI, texture, annotation overlay or external dataset. Test each delivery path separately.
Protect glTF buffer layout and accessors
Mesh geometry and related data may be addressed through bufferViews and accessors. Those offsets, component types and counts must remain valid.
Worked example: A product team stores a building mesh in a GLB. Editing a label outside the binary buffer should not change buffer byte offsets for positions and normals.
Failure to avoid: Using a plain text editor on binary payloads may corrupt alignment or lengths. A loader error can occur far from the attempted language change.
Validation: Validate glTF payloads independently and compare structural and geometric measures to the source. Avoid unsupported byte-level rewrites.
Respect feature identity in metadata
Metadata can connect individual rendered features to stable IDs and typed properties. Translating the displayed property name does not mean renumbering the feature.
Worked example: An inspector clicks building 372 and sees its maintenance status and localized description. The same building must resolve to the same record in every language.
Failure to avoid: If feature IDs shift, a viewer can show correct-looking geometry while retrieving maintenance data for a different object.
Validation: Compare feature counts, IDs and property-table associations. Test selection of several known features against their source records.
Know when EXT_mesh_features is relevant
glTF’s feature metadata mechanisms can identify parts of meshes for attribute lookup. The exact extension and supported semantics depend on the payload and implementation.
Worked example: A photogrammetric city mesh contains selectable surfaces with feature identities. The language layer should attach descriptions to stable entities rather than to triangle order.
Failure to avoid: Treating mesh-feature IDs as text labels or editing feature-ID attributes during translation can break picking and analytics.
Validation: Inspect extension declarations, feature-ID mappings and the viewer’s metadata support. Preserve identifiers and validate selection after localized labels appear.
Keep EXT_structural_metadata schema semantics
Typed property tables and metadata schemas distinguish property identities and datatypes from human descriptions. The schema serves software; descriptive labels serve people.
Worked example: An urban asset property stores floorCount as an integer and an optional localized explanatory label. The numeric property must remain machine-accessible.
Failure to avoid: Renaming property identifiers for each language prevents shared queries and dashboards from working across editions. Changing numeric types to strings is equally risky.
Validation: Preserve schema identifiers and types; localize approved display terms through a documented lookup or client presentation layer.
Treat metadata values according to type
An attribute may contain a string, numeric measurement, enumeration, boolean or reference. Only some string fields are genuinely translatable prose.
Worked example: A road-feature record includes a category code, a surface area and a free-text public note. Only the note is a straightforward translation target.
Failure to avoid: Applying language software to all values turns 12.5 into a locale-formatted string or transforms a controlled status code into an unsupported word.
Validation: Use the metadata schema to separate language fields from typed or controlled data. Compare invariant values programmatically.
Keep stable selection and picking behavior
A viewer uses feature references to connect points on the screen to data records. The same object must be selected regardless of language.
Worked example: A visitor selects the same bridge in English and Arabic and expects the same structure, description and accessibility status.
Failure to avoid: If the localized variant swaps feature metadata offsets, a popup can describe a neighboring bridge. The scene may render perfectly while conveying false information.
Validation: Run a click-selection matrix across several known objects in every supported language and compare stable IDs.
Design multilingual name and description tables
Stable feature IDs can index a table of translated display names and descriptions. This avoids modifying the geometry or standard metadata identifiers for each locale.
Worked example: A cultural-site application stores English, Malay and Japanese labels against one feature ID and chooses the appropriate label at runtime.
Failure to avoid: Duplicating entire tilesets per language without governance increases storage and creates drift when the canonical geometry changes.
Validation: Keep one authoritative feature identity, versioned language strings and a defined fallback order. Document which name variants are official.
Avoid translating technical enum values
Values that control rendering, classifications or filters are not equivalent to interface strings. They may look like readable English but follow a defined vocabulary.
Worked example: A property identifies a feature type used to color all public facilities. The user can see a translated legend while the internal code remains constant.
Failure to avoid: A translated enum value may silently break a style expression or filter. The viewer then stops grouping objects as expected.
Validation: Test style and filter queries before and after adding localized labels and ensure machine values are preserved.
Read declarative styles as instructions
3D Tiles styles can use expressions to set show and color based on feature properties. They are executable presentation rules, not ordinary sentences.
Worked example: An application colors buildings by measured height. An Arabic legend can explain the colors without changing the expression used to compare numeric Height values.
Failure to avoid: Blindly translating property references inside an expression can make it fail evaluation, even when the new sentence is grammatically excellent.
Validation: Check every expression reference against the metadata schema. Compare visual classification at known test features across languages.
Protect color meaning and warning semantics
A color assignment may encode warnings, occupancy categories or accessibility information. The translated legend must agree with that established meaning.
Worked example: A city operations map uses amber for inspection due and red for a restriction. Translating the legend must not imply that amber means urgent closure.
Failure to avoid: Changing symbolic meanings to suit a culture without checking operational policy can misinform users. Accessibility also requires information beyond color alone.
Validation: Review legends with the data authority and test text equivalents, contrast and screen-reader access for important status information.
Understand implicit tiling and subtrees
Implicit tiling compactly describes repetitive spatial hierarchies through subdivision schemes and subtree availability data. These are structural rules rather than prose.
Worked example: A global city scene uses an octree-style implicit hierarchy to reduce metadata overhead. Its local-language presentation changes at a popup layer, not within availability bits.
Failure to avoid: Rewriting subtree file names or availability arrays because their encoding looks cryptic can cause unpredictable missing tiles at depth.
Validation: Validate subtree structures and compare availability at selected coordinates and levels in the source and target releases.
Preserve availability bitstreams
Subtree availability encodes which tiles and content are present. The bits are data used for streaming requests; their order is not a translation choice.
Worked example: A district has empty branches where imagery is absent. A viewer should not request nonexistent tiles because a translated package accidentally changed availability information.
Failure to avoid: A mismatched availability bitstream can prevent legitimate content from loading or trigger requests for absent assets. Failure may appear only after prolonged navigation.
Validation: Compare availability data, target URIs and subtree integrity. Test high- and low-detail viewpoints beyond the initial camera location.
Distinguish nested tilesets from tile contents
A tile may point to a separate tileset hierarchy instead of a renderable mesh. That reference expands the scene structure for a client.
Worked example: A national model links regional tilesets that themselves link neighborhood assets. Translating a regional display name must not break these nested references.
Failure to avoid: Treating a nested tileset URI as an ordinary asset label can cut off all descendants under the linked region.
Validation: Traverse and validate external tileset links recursively, including permissions and relative URI resolution.
Inspect asset-relative paths and deployment URLs
A packaged tileset can rely on relative URLs whose meaning depends on the publishing directory. Language-specific web routes may alter that base path.
Worked example: A localized site publishes the same tileset under a different URL prefix. Relative texture files that previously loaded may now return 404 responses.
Failure to avoid: A route change can damage resource loading without changing JSON contents. Fragment identifiers, encoding and signed URLs add additional complexity.
Validation: Test network requests for models, textures, metadata schemas and subtrees in the actual deployment environment, not only on a local machine.
Preserve compression and extension requirements
Some tile payloads use extensions or compression that require viewer support. The JSON extension declarations and binary encodings are technical requirements.
Worked example: A model uses compressed geometry for efficient mobile streaming. Its multilingual web wrapper must still load that content with a compatible engine.
Failure to avoid: Removing an extension declaration to make a file look simpler may produce loader failures or wrong geometry interpretation.
Validation: Record required extensions and test with the exact target viewers. Keep source payloads immutable through the language workflow.
Understand legacy b3dm i3dm pnts and cmpt
Earlier 3D Tiles datasets may contain Batched 3D Model, Instanced 3D Model, Point Cloud and Composite tile formats. They are legacy but still found.
Worked example: An archive from a city project contains b3dm tiles while a new release uses glTF-oriented 1.1 content. Localization guidance must distinguish them.
Failure to avoid: Calling an old format simply obsolete and rewriting it by name is not a conversion strategy. A migration may change metadata organization and asset behavior.
Validation: Identify content magic and actual payload types. Keep older datasets compatible or run an explicitly tested technical migration separate from translation.
Write accurate multilingual camera-facing UI labels
Viewer controls, information panels, search results and tour instructions are often the most visible translation targets, but they may live outside the tileset.
Worked example: A public exhibition shows buttons for Reset view, Show buildings and Read history. Translating those controls does not require modifying bounding volumes or glTF buffers.
Failure to avoid: A purely file-centric process can waste effort touching protected fields while leaving visitor controls untranslated or inaccessible.
Validation: Inventory the complete user journey and localize interface strings through the application’s normal internationalization mechanism.
Keep units and measurements unambiguous
Height, floor area, distance and geographic position may appear in captions. Their underlying numeric metadata and units must remain defined independently of language.
Worked example: A viewer shows a 42-meter observation deck and a locale may display an additional feet conversion. Both readers should refer to the same physical height.
Failure to avoid: Formatting 42.0 as a comma-decimal string in the metadata may change its datatype. Unlabeled conversions can make a building look taller than its source.
Validation: Preserve canonical numeric values and declared units. Verify localized display conversions and precision with a known test case.
Protect time-dependent attributes
Some viewers attach inspection dates, construction eras or changing sensor values to features. A translated status must preserve the time scope of the claim.
Worked example: A digital twin dashboard says an elevator was inspected last month and displays the date according to the viewer’s locale. The underlying instant must remain the same.
Failure to avoid: An ambiguous date translation can change chronology. An old snapshot may be mistaken for live status when language labels omit temporal context.
Validation: Display time zones and update dates clearly. Compare timestamps and source provenance before accepting time-related wording.
Avoid creating false authority in named places
Building names, institution labels and historical descriptions may be official, colloquial, disputed or newly proposed. Translation should preserve those distinctions.
Worked example: A redevelopment proposal displays a planned plaza name. A target-language viewer should not present that proposed name as an existing legally approved place.
Failure to avoid: Replacing an official proper noun with a literal translation can break search and suggest an authority that does not exist.
Validation: Review approved name registers, mark provisional translations and retain canonical identifiers in search indexes.
Handle text drawn inside image textures carefully
Localized signage embedded in facade textures requires controlled artistic editing or alternate imagery while protecting non-text visual evidence.
Worked example: A museum scene has a directional sign embedded in a texture atlas. A translated texture can be created as a separate asset with matching dimensions and UV mapping.
Failure to avoid: Painting over a texture may remove architectural details or cause the label to appear on the wrong object if the atlas layout changes.
Validation: Test localized texture placement from several camera angles and confirm the underlying geometry and feature IDs have not been altered.
Respect intellectual property and asset permissions
A tileset can combine photogrammetry, private imagery, licensed building models and third-party textures. Language access does not automatically grant republication rights.
Worked example: A consortium permits translated descriptive metadata but restricts redistribution of the underlying 3D capture. The release must honor both conditions.
Failure to avoid: Packaging all source assets into an unrestricted multilingual download can exceed the rights provided by the original license.
Validation: Check permitted use for each input asset and record public access boundaries before distributing a localized scene.
Design an independent validation matrix
Structure, performance, language quality and accessibility are separate dimensions. A single screenshot is evidence for none of them in full.
Worked example: A team tests two languages, three viewer types, desktop and mobile, several levels of detail and object picking around known features.
Failure to avoid: A file that opens once can still fail deep zoom, right-to-left rendering, cached resource resolution or missing language fallback.
Validation: Record pass/fail outcomes for structural validation, metadata invariants, network requests, tile selection, selection accuracy and language review.
Use reproducible visual comparison positions
Fixed camera parameters make source and target screenshots meaningfully comparable. Otherwise different viewpoints can hide missing tiles or classification changes.
Worked example: Reviewers inspect a bridge, an industrial district and a dense downtown block from saved camera poses. The localized descriptions differ, but geometry should align.
Failure to avoid: Visual inspection from random viewpoints can miss omissions. An asset may load only at a specific zoom range or camera heading.
Validation: Store consistent camera locations, viewport size and rendering settings when comparing editions, then separately test exploratory navigation.
Audit data privacy and security-sensitive layers
Large 3D scenes may reveal infrastructure, access details or private information. A new language edition may expand who can discover that data.
Worked example: An internal facilities twin includes access pathways. Translating captions for a public portal is inappropriate unless the underlying release scope is approved.
Failure to avoid: The main risk can be publication rather than translation accuracy. Making hidden facility metadata searchable in more languages may widen disclosure.
Validation: Perform an access and disclosure review independently of linguistic correctness. Publish only assets and attributes approved for the intended audience.
Plan fallbacks instead of inventing strings
When an approved translation is missing, viewers need a predictable fallback language or a clear unavailable indicator. They should not manufacture official-sounding descriptions.
Worked example: A tourist requests a locale with only partial coverage. The viewer shows the established source-language name and an explanatory language marker.
Failure to avoid: Automatic fallback that silently replaces a formal asset name with an unreviewed translation can create confusion about identity or legal status.
Validation: Test missing-language and partial-language scenes, including search and screen readers, and make fallback provenance visible.
Release a verifiable language edition
A credible release keeps a canonical tileset, a map of authorized language changes and evidence that the scene remains structurally equivalent.
Worked example: A city publishes one geometry dataset and three localized presentation bundles. Reviewers can compare feature IDs and tile assets even after text updates.
Failure to avoid: Duplicating datasets per language without controls can cause later geometry corrections to reach one edition but not another.
Validation: Maintain a versioned manifest with source asset hashes, changed string fields, reviewer approvals and validation results.
A twelve-step procedure for an international 3D Tiles release
Step 1. Identify the actual tileset and its declared specification version. Keep an immutable copy of tileset.json and enumerate nested tilesets, content types, glTF extensions, schemas, subtrees and external asset references before any localization edit.
Step 2. Create a rendering baseline using fixed camera positions at overview, neighborhood and building scales. Record feature picking results, loading performance, missing tiles and the application versions used for the comparison.
Step 3. Separate language fields into interface text, feature-property labels, human descriptions, baked texture text and official place names. Treat each category differently because its ownership and technical dependency are different.
Step 4. Write a field-level contract that specifies which strings can change and which identifiers, types, paths, coordinate values, matrices, geometry and metadata schema elements cannot. Any deviation needs explicit technical approval.
Step 5. Compile an approved multilingual terminology list with a source-language record, official name policy, transliteration decisions, technical definitions and the owner responsible for accepting each translated value.
Step 6. Select the safest delivery arrangement. Prefer stable canonical geometry with sidecar localization tables keyed by feature ID or client internationalization files where the application supports those approaches.
Step 7. Translate a small scene first. Include one selectable building, one model texture with text, one style legend, one nested tile and one known fallback to exercise the range of language surfaces.
Step 8. Validate structural integrity before visual inspection. Parse tileset JSON, validate glTF assets, check all referenced resources and compare source-to-target tree properties, feature identifiers and typed metadata.
Step 9. Test the scene from the same saved camera positions and through dynamic navigation. Look for disappearing regions, wrong refinement, misplaced geometry, missing textures and broken picking in each language edition.
Step 10. Review target-language use with people familiar with the place and domain. Confirm proper names, administrative status, safety messages, measurement units and historical descriptions independently of JSON validity.
Step 11. Check accessibility, user search and partial-language fallbacks on both desktop and mobile devices. Validate right-to-left presentation and mixed-script rendering when relevant to the target audiences.
Step 12. Publish the localization with a clear version and release receipt. Document the exact string changes, source dataset, unchanged technical contracts, tested viewers and any remaining known limitations.
Frequently asked questions about translating 3D Tiles
What is 3D Tiles 1.1 used for?
It is an open OGC community standard for efficiently streaming large-scale three-dimensional geospatial content to applications. A tileset uses a spatial hierarchy, bounding information, tile content references and metadata. It can be used for buildings, urban digital twins, photogrammetry and point clouds. Translating the surrounding user experience should not modify the geospatial content needed by rendering and analysis software.
Can I translate tileset.json property names?
No. Properties such as root, children, geometricError, boundingVolume and content are structural syntax. Changing them to another natural language breaks compatibility. A label, description or explicitly supported locale-specific metadata field may be translated; the standard’s machine keys and controlled values should stay stable.
Should each language have its own complete tileset?
That is a deployment choice, not a requirement. Many applications benefit from keeping the canonical tileset and geometry unchanged while storing multilingual labels in a separate keyed data layer. Separate full tilesets may be necessary for localized texture images or delivery restrictions, but then asset versions and geometry must be reconciled carefully to prevent drift.
What does geometricError mean?
It is a numeric measure used in tile selection to control the level of detail a viewer requests. It is not a translation quality score. Changing the value may cause a model to load too coarsely or too aggressively. Explanations of detail level can be translated in the interface without changing the underlying tile-selection mathematics.
How are glTF feature names different from feature IDs?
A human-readable name can have different language editions. A feature ID is a stable technical reference used to connect geometry to typed metadata, picking and external records. Translating or renumbering an ID to match a display name risks associating the wrong records with the wrong rendered objects.
What if a word is embedded in a building texture?
It may require a separately produced texture variation. Protect the original geometry, UV mapping, material relationships and asset rights; verify that translated writing appears in the correct physical location. This is different from editing a text string in a popup and should be reviewed by a 3D asset specialist.
Are 3D Tiles and Mapbox Vector Tiles the same?
No. Both organize data for efficient viewing, but Mapbox Vector Tiles concern encoded 2D vector features and their rendering, while 3D Tiles specializes in streamed three-dimensional content with tile hierarchy, glTF assets, spatial volumes and detailed scene selection. Their technical invariants and localization points are different.
How do I prove a translated 3D scene still represents the original?
Use multiple kinds of evidence: unchanged tile hierarchy and structural properties, valid referenced resources, geometry and feature-ID comparisons, fixed-camera visual tests, known-object picking, performance checks and native-speaker language review. Keep a release record identifying what did change. A single attractive screenshot cannot prove full equivalence.
Conclusion: a translated interface must still render the same world
A reliable multilingual 3D Tiles edition makes large 3D places intelligible without making them different. Keep tile boundaries, selection thresholds, matrices, content URIs, glTF geometry, feature IDs and technical metadata stable. Translate approved human descriptions, map legends, proper-name displays and user controls. Validate the language in context and test the streaming scene from many viewpoints. The user gains access to meaning while the 3D system retains its original evidence.
Standards and nearby technical guides
OGC 3D Tiles 1.1 specification; OGC 3D Tiles standards overview; Khronos glTF specification and resources; Companion: semantic 3D city models in CityGML 3.0; Companion: Mapbox 2D vector tiles.