-
Notifications
You must be signed in to change notification settings - Fork 292
XKT Format V4
See also:
- XKTLoaderPlugin - loading models from
.xkt
format- Creating Files for Offline BIM - converting IFC, OBJ and glTF files to
.xkt
format
The .xkt
format is xeokit's native binary format, optimized to allow large IFC and CAD models to be efficiently loaded into
xeokit using the XKTLoaderPlugin.
At present, the .xkt
only supports triangle meshes.
See the tutorial Creating Files for Offline BIM for how to use existing tools to convert IFC, OBJ, and glTF files to .xkt
format.
As described in that tutorial, we currently have xeokit-gltf-to-xkt, a nodejs-based CLI tool that converts glTF
files to .xkt
. That tool performs well and should be enough for your needs.
This specification page describes the .xkt
V4.0 format, in case you need to develop own custom conversion tool, for example on IFC Engine.
In that case, use the source code for xeokit-gltf-to-xkt as a reference implementation to help you understand this specification.
- XKT V4.0
- Overview
- zlib Deflation
- Geometry Arrays
- Implicit Mesh Order
- Indices
- Primitives
- Entity IDs
- Terminology
The .xkt
V4 format organizes model geometry into a minimal payload that can be loaded efficiently over the Web.
The format minimizes the size of the model using the following techniques:
- Quantize 32-bit geometry vertex positions to unsigned 16-bit integers,
- oct-encode 32-bit vertex normals to unsigned 8-bit integers, and
- deflate everything using zlib.
To avoid loss of precision after quantization, the xeokit-gltf-to-xkt tool uses a k-d tree to partition the vertex positions into sub-regions, which are each quantized separately into the full 16-bit range.
The table below lists the elements within .xkt
V4.0. For convence we're using a symbolic name (Eg. size_index
) for each element.
Elements deflated with zlib are flagged in the fourth column.
Element | Type | Description | zlib Deflated? |
---|---|---|---|
version |
Uint32 | The .xkt file format version. This is the first four bytes in the file, and will have the value 4 . |
|
size_index |
Uint32 | Byte size of the index. The index is the following block of elements that are prefixed with size_ . The index provides a table of the sizes of elements within the file. |
|
size_positions |
Uint32 | Byte size of deflated positions . This is the start of the index. |
|
size_normals |
Uint32 | Byte size of deflated normals . |
|
size_indices |
Uint32 | Byte size of deflated indices . |
|
size_edge_indices |
Uint32 | Byte size of deflated edge_indices . |
|
size_decode_matrices |
Uint32 | Byte size of deflated decode_matrices . |
|
size_each_primitive_positions_and_normals_portion |
Uint32 | Byte size of deflated each_primitive_positions_and_normals_portion . |
|
size_each_primitive_indices_portion |
Uint32 | Byte size of deflated each_primitive_indices_portion . |
|
size_each_primitive_edge_indices_portion |
Uint32 | Byte size of deflated each_primitive_edge_indices_portion . |
|
size_each_primitive_decode_matrices_portion |
Uint32 | Byte size of deflated each_primitive_decode_matrices_portion . |
|
size_each_primitive_color |
Uint32 | Byte size of deflated each_primitive_color . |
|
size_primitive_instances |
Uint32 | Byte size of deflated primitive_instances . |
|
size_each_entity_id |
Uint32 | Byte size of deflated each_entity_id . |
|
size_each_entity_primitive_instances_portion |
Uint32 | Byte size of deflated each_entity_primitive_instances_portion . |
|
size_each_entity_matrix |
Uint32 | Byte size of deflated each_entity_matrix . |
|
positions |
Uint16[] | Quantized vertex positions for all primitives. Each mesh has a portion of this array. Portions for primitives that are only used by one entity are in World Space coordinates. Portions that are used by multiple primitives are in Model Space. | Deflated |
normals |
Uint8[] | Oct-encoded vertex normals for all primitives. Each mesh has a portion of this array. | Deflated |
indices |
Uint32[] | Geometry triangle indices for all primitives. Has three elements per triangle. Each mesh has a portion of this array. | Deflated |
edge_indices |
Uint32[] | Geometry edge indices for all primitives. Has two elements per edge. Each mesh has a portion of this array. | Deflated |
decode_matrices |
Float32[] | Decode matrices to de-quantize positions . Each of these corresponds to a portion of positions . These may be shared by multiple primitives. |
Deflated |
each_primitive_positions_and_normals_portion |
Uint32[] | For each mesh, base index of a portion in positions and normals . |
Deflated |
each_primitive_indices_portion |
Uint32[] | For each mesh, base index of a portion in indices . |
Deflated |
each_primitive_edge_indices_portion |
Uint32[] | For each mesh, base index of a portion in edge_indices . |
Deflated |
each_primitive_decode_matrices_portion |
Uint32[] | For each mesh, base index of a portion in decode_matrices . Primitives can share decode matrices. |
Deflated |
each_primitive_color |
Uint8[] | For each mesh, an RGBA color. Has four elements per color, each in range [0..255] . The fourth element, alpha, is opacity. |
Deflated |
primitive_instances |
Uint32[] | For each mesh instance, the base index of the primitive's portion of the each_primitive_* arrays. |
Deflated |
each_entity_id |
String | ID for each entity. This is a string-encoded JSON array of strings. | Deflated |
each_entity_primitive_instances_portion |
Float32[] | For each entity, base index of the entity's portion in primitive_instances . |
Deflated |
each_entity_matrix |
Float32[] | For each entity, a 16-element modeling transform matrix. | Deflated |
Note the last column in the table above, which indicates that some of the elements are deflated using zlib. The xeokit-gltf-to-xkt tool and the XKTLoaderPlugin plugin both use pako.js, which is a JavaScript port of zlib, to deflate and inflate.
When loading .xkt
, XKTLoaderPlugin
inflates those elements before parsing them.
The positions
, normals
, indices
and edge_indices
arrays are the concatenation of the geometries for all the primitives in the model.
Primitives can be shared by multiple entities. For a mesh that's used by only one entity, its positions
portion will be in World Space. For a mesh that's used by multiple entities, its positions
portion will be in Model Space.
The positions
array is quantized to 16-bit integers, and will be dequantized in xeokit's shaders using a corresponding decode matrix in decode_matrices
.
The normals
array is oct-encoded to 8-bit integers, and will be also decoded in xeokit's shaders (no matrix is used for oct-decoding).
For an example of geometry quantization and oct-encoding using JavaScript and WebGL, see the mesh-quantization-example demo by @tsherif. You can also find an example within the source code of xeokit-gltf-to-xkt.
The indices
array defines triangles, with three elements per triangle.
The edge_indices
array defines the edges that xeokit draws for wireframe views, with two elements per edge. An .xkt
exporter needs to generate those edge indices from the geometries, using the algorithm demonstrated in buildEdgesindices.js (a file within the xeokit-gltf-to-xkt converter tool). Edge generation is computationally expensive. It's cheaper to pre-generate them in the .xkt
file, rather than have XKTLoaderPlugin generate them on-the-fly while loading.
The each_primitive_positions_and_normals_portion
, each_primitive_indices_portion
and each_primitive_edge_indices_portion
indicate which portion of the geometry arrays is
used for each mesh.
The first vertex position used by mesh meshIdx
is:
let i = each_primitive_positions_and_normals_portion[ meshIdx ];
let x = positions[ i + 0 ];
let y = positions[ i + 1 ];
let z = positions[ i + 2 ];
The last vertex position used by mesh meshIdx
is:
let i2 = each_primitive_positions_and_normals_portion[ meshIdx + 1 ] - 1;
let x2 = positions[ i2 + 0 ];
let y2 = positions[ i2 + 1 ];
let z2 = positions[ i2 + 2 ];
The first vertex normal used by mesh meshIdx
is:
let i = each_primitive_positions_and_normals_portion[ meshIdx ];
let x = normals[ i + 0 ];
let y = normals[ i + 1 ];
let z = normals[ i + 2 ];
The last vertex normal used by mesh meshIdx
is:
let i2 = each_primitive_positions_and_normals_portion[ meshIdx + 1 ] - 1;
let x2 = normals[ i2 + 0 ];
let y2 = normals[ i2 + 1 ];
let z2 = normals[ i2 + 2 ];
Recall that positions
are quantized to 16-bit integers and normals
are oct-encoded. To de-quantize positions
back to floating point values, xeokit will multiply them by the corresponding decode matrix in decode_matrices
.
The indices
array indexes positions
and normals
to define the geometry primitives, which are triangles.
Like positions
and normals
, indices
has a portion for each mesh.
In the snippet below, we'll obtain the quantized positions of the vertices of the first triangle for mesh meshIdx
:
let indicesBaseIdx = each_primitive_indices_portion[ meshIdx ];
let positionsAndNormalsBaseIdx = each_primitive_positions_and_normals_portion[ meshIdx ];
let a = indices[ indicesBaseIdx + 0 ];
let b = indices[ indicesBaseIdx + 1 ];
let c = indices[ indicesBaseIdx + 2 ];
let ax = positions[ positionsAndNormalsBaseIdx + (a * 3) + 0];
let ay = positions[ positionsAndNormalsBaseIdx + (a * 3) + 1];
let az = positions[ positionsAndNormalsBaseIdx + (a * 3) + 2];
let bx = positions[ positionsAndNormalsBaseIdx + (b * 3) + 0];
let by = positions[ positionsAndNormalsBaseIdx + (b * 3) + 1];
let bz = positions[ positionsAndNormalsBaseIdx + (b * 3) + 2];
let cx = positions[ positionsAndNormalsBaseIdx + (c * 3) + 0];
let cy = positions[ positionsAndNormalsBaseIdx + (c * 3) + 1];
let cz = positions[ positionsAndNormalsBaseIdx + (c * 3) + 2];
Similarly, we'll obtain the oct-encoded normals for the vertices of that triangle:
let anx = normals[ positionsAndNormalsBaseIdx + (a * 3) + 0];
let any = normals[ positionsAndNormalsBaseIdx + (a * 3) + 1];
let anz = normals[ positionsAndNormalsBaseIdx + (a * 3) + 2];
let bnx = normals[ positionsAndNormalsBaseIdx + (b * 3) + 0];
let bny = normals[ positionsAndNormalsBaseIdx + (b * 3) + 1];
let bnz = normals[ positionsAndNormalsBaseIdx + (b * 3) + 2];
let cnx = normals[ positionsAndNormalsBaseIdx + (c * 3) + 0];
let cny = normals[ positionsAndNormalsBaseIdx + (c * 3) + 1];
let cnz = normals[ positionsAndNormalsBaseIdx + (c * 3) + 2];
Note how each_primitive_indices_portion
contains a base index for each mesh to indicate its portion of indices
, and each_primitive_positions_and_normals_portion
contains a base index for each mesh to indicate its portion of positions
. We use each_primitive_positions_and_normals_portion
to offset each
index to align it with the primitives portion in positions
.
In xeokit, an entity can have multiple primitives. For example, an entity representing a window could have a mesh representing the frame, another representing the pane, another for the handle, and so on.
The primitive_instances
array contains a base index into each_primitive_positions_and_normals_portion
, primitive_normals
,
each_primitive_indices_portion
and each_primitive_color
for each entity.
Let's extend the previous snippet to obtain the quantized positions and oct-encoded normals of the vertices of the first triangle within the first
mesh belonging to the entity at entityIdx
:
let meshBaseIdx = each_entity_primitive_instances_portion[ entityIdx ];
let indicesBaseIdx = each_primitive_indices_portion[ meshBaseIdx ];
let positionsAndNormalsBaseIdx = each_primitive_positions_and_normals_portion[ meshBaseIdx ];
let a = indices[ indicesBaseIdx + 0 ];
let b = indices[ indicesBaseIdx + 1 ];
let c = indices[ indicesBaseIdx + 2 ];
let ax = positions[ positionsAndNormalsBaseIdx + (a * 3) + 0];
let ay = positions[ positionsAndNormalsBaseIdx + (a * 3) + 1];
let az = positions[ positionsAndNormalsBaseIdx + (a * 3) + 2];
let bx = positions[ positionsAndNormalsBaseIdx + (b * 3) + 0];
let by = positions[ positionsAndNormalsBaseIdx + (b * 3) + 1];
let bz = positions[ positionsAndNormalsBaseIdx + (b * 3) + 2];
let cx = positions[ positionsAndNormalsBaseIdx + (c * 3) + 0];
let cy = positions[ positionsAndNormalsBaseIdx + (c * 3) + 1];
let cz = positions[ positionsAndNormalsBaseIdx + (c * 3) + 2];
let anx = normals[ positionsAndNormalsBaseIdx + (a * 3) + 0];
let any = normals[ positionsAndNormalsBaseIdx + (a * 3) + 1];
let anz = normals[ positionsAndNormalsBaseIdx + (a * 3) + 2];
let bnx = normals[ positionsAndNormalsBaseIdx + (b * 3) + 0];
let bny = normals[ positionsAndNormalsBaseIdx + (b * 3) + 1];
let bnz = normals[ positionsAndNormalsBaseIdx + (b * 3) + 2];
let cnx = normals[ positionsAndNormalsBaseIdx + (c * 3) + 0];
let cny = normals[ positionsAndNormalsBaseIdx + (c * 3) + 1];
let cnz = normals[ positionsAndNormalsBaseIdx + (c * 3) + 2];
Recall that primitive_instances
has the number of primitives used by each entity. To get the base index for the last mesh of our entity:
let numPrimitivesInEntity = primitive_instances[ entityIdx];
let meshBaseIdx = each_entity_primitive_instances_portion[ entityIdx + numPrimitivesInEntity - 1 ];
//...
xeokit uses the decode matrices in decode_matrices
to convert quantized positions
back to 32-bit floating point values. xeokit uses those matrices on the GPU, within its shaders.
Each mesh has a decode matrix in decode_matrices
. A decode matrix can be shared by multiple primitives.
The each_primitive_decode_matrices_portion
array contains a base index into decode_matrices
for each mesh.
Let's get the decode matrix for the first mesh belonging to the entity at entityIdx
:
let meshBaseIdx = each_entity_primitive_instances_portion[ entityIdx ];
let deocdeMatrixBaseIdx = each_primitive_decode_matrices_portion[ meshBaseIdx ];
let decodeMatrix = Float32Array(16);
for (let i = 0; i < 16; i++) {
decodeMatrix[i] = decode_matrices[ deocdeMatrixBaseIdx + i];
}
As mentioned earlier, .xkt
reduces accuracy loss by partitioning the positions
into sub-regions, which are quantized separately. Each of the decode matrices corresponds to one of these partitions. Portions of positions
that fall inside the same partition will share the same decode matrix.
Each entity has a string ID, which we can get like so:
let entityId = each_entity_id[ entityIdx ];
A mesh represents an element of geometry. Each mesh has its geometry in a portion of the positions
, normals
, indices
and edgeIndices
arrays.
Primitives are used by entities. Some primitives are used by exactly one entity, with others are shared by multiple entities. When shared, we say that a mesh is instanced by the entities that share it. Instancing is very desirable in our models, because it reduces file size.
A mesh that's only used by one entity has positions
that are in World-space. A mesh used by multiple entities has positions
in Model-space.
For shared primitives, xeokit will use the modeling matrices on their entities to transform their Model-space positions
into World-space.
An entity is a drawable element within a xeokit model. An entity is comprised of one or more primitives.
TODO
TODO
Quantization is the process of converting 32-bit floating point values for vertex positions
into unsigned 16-bit integers. This reduces the space occupied by each vertex position from 12 bytes down to six bytes, which reduces the size of the XKT file.
Quantization is performed by scaling and translating the floating point values so that they fit inside the unsigned 16-bit range. The quantized positions
are accompanied by "decoding matrices" in decodeMatrices
, that contain the scaling and translation that xeokit performs on the GPU to de-quantize them back to their original 32-bit floating-point values.
Some loss of precision occurs during the conversion just mentioned. For this reason, XKT V4 divides the positions
into sub-regions, with each sub-region quantized independently into the full 16-bit range. Each sub-region has its own decoding matrix.
Resources:
- The xeokit-gltf-to-xkt tool contains quantization functions and uses a k-d tree to partition the
positions
into sub-regions. - See mesh-quantization-example for a WebGL-based quantization demo.
- Read the Cesium blog article Using Quantization with 3D Models for more information on quantization.
Oct Encoding is the process of converting 32-bit floating point values for vertex normals
into unsigned 8-bit integers. As with quantization, this reduces the space they occupy. Similarly, xeokit converts them back to 32-bit flots on the GPU.
Oct-encoded normals don't need a decoding matrix for xeokit to convert them back to floats.
Resources:
- The xeokit-gltf-to-xkt tool contains oct-encoding functions.
- See mesh-quantization-example for a WebGL-based oct-encoding demo.