Assimp  v3.1.1 (June 2014)
pyassimp.postprocess Namespace Reference

Variables

int aiProcess_CalcTangentSpace = 0x1
 
tuple aiProcess_ConvertToLeftHanded
 
int aiProcess_Debone = 0x4000000
 
int aiProcess_FindDegenerates = 0x10000
 
int aiProcess_FindInstances = 0x100000
 
int aiProcess_FindInvalidData = 0x20000
 
int aiProcess_FixInfacingNormals = 0x2000
 
int aiProcess_FixTexturePaths = 0x200000
 
int aiProcess_FlipUVs = 0x800000
 
int aiProcess_FlipWindingOrder = 0x1000000
 
int aiProcess_GenEntityMeshes = 0x100000
 
int aiProcess_GenNormals = 0x20
 
int aiProcess_GenSmoothNormals = 0x40
 
int aiProcess_GenUVCoords = 0x40000
 
int aiProcess_ImproveCacheLocality = 0x800
 
int aiProcess_JoinIdenticalVertices = 0x2
 
int aiProcess_LimitBoneWeights = 0x200
 
int aiProcess_MakeLeftHanded = 0x4
 
int aiProcess_OptimizeAnimations = 0x200000
 
int aiProcess_OptimizeGraph = 0x400000
 
int aiProcess_OptimizeMeshes = 0x200000
 
int aiProcess_PreTransformVertices = 0x100
 
int aiProcess_RemoveComponent = 0x10
 
int aiProcess_RemoveRedundantMaterials = 0x1000
 
int aiProcess_SortByPType = 0x8000
 
int aiProcess_SplitByBoneCount = 0x2000000
 
int aiProcess_SplitLargeMeshes = 0x80
 
int aiProcess_TransformUVCoords = 0x80000
 
int aiProcess_Triangulate = 0x8
 
int aiProcess_ValidateDataStructure = 0x400
 
tuple aiProcessPreset_TargetRealtime_Fast
 
tuple aiProcessPreset_TargetRealtime_MaxQuality
 
tuple aiProcessPreset_TargetRealtime_Quality
 

Variable Documentation

int pyassimp.postprocess.aiProcess_CalcTangentSpace = 0x1
tuple pyassimp.postprocess.aiProcess_ConvertToLeftHanded
Initial value:
1 = ( \
2  aiProcess_MakeLeftHanded | \
3  aiProcess_FlipUVs | \
4  aiProcess_FlipWindingOrder | \
5  0 )
int pyassimp.postprocess.aiProcess_Debone = 0x4000000

This step removes bones losslessly or according to some threshold.

In some cases (i.e. formats that require it) exporters are forced to assign dummy bone weights to otherwise static meshes assigned to animated meshes. Full, weight-based skinning is expensive while animating nodes is extremely cheap, so this step is offered to clean up the data in that regard.

Use AI_CONFIG_PP_DB_THRESHOLD<tt> to control this. Use AI_CONFIG_PP_DB_ALL_OR_NONE<tt> if you want bones removed if and only if all bones within the scene qualify for removal.

int pyassimp.postprocess.aiProcess_FindDegenerates = 0x10000

This step searches all meshes for degenerate primitives and converts them to proper lines or points.

A face is 'degenerate' if one or more of its points are identical. To have the degenerate stuff not only detected and collapsed but removed, try one of the following procedures:
1. (if you support lines and points for rendering but don't want the degenerates)

  • Specify the aiProcess_FindDegenerates flag.
  • Set the AI_CONFIG_PP_FD_REMOVE option to 1. This will cause the step to remove degenerate triangles from the import as soon as they're detected. They won't pass any further pipeline steps.

    2.(if you don't support lines and points at all)
    • Specify the aiProcess_FindDegenerates flag.
    • Specify the aiProcess_SortByPType flag. This moves line and point primitives to separate meshes.
    • Set the AI_CONFIG_PP_SBP_REMOVE option to
      1 aiPrimitiveType_POINTS | aiPrimitiveType_LINES
      to cause SortByPType to reject point and line meshes from the scene.
      Degenerate polygons are not necessarily evil and that's why they're not removed by default. There are several file formats which don't support lines or points, and some exporters bypass the format specification and write them as degenerate triangles instead.

int pyassimp.postprocess.aiProcess_FindInstances = 0x100000

This step searches for duplicate meshes and replaces them with references to the first mesh.

This step takes a while, so don't use it if speed is a concern. Its main purpose is to workaround the fact that many export file formats don't support instanced meshes, so exporters need to duplicate meshes. This step removes the duplicates again. Please note that Assimp does not currently support per-node material assignment to meshes, which means that identical meshes with different materials are currently #not# joined, although this is planned for future versions.

int pyassimp.postprocess.aiProcess_FindInvalidData = 0x20000

This step searches all meshes for invalid data, such as zeroed normal vectors or invalid UV coords and removesfixes them. This is intended to get rid of some common exporter errors.

This is especially useful for normals. If they are invalid, and the step recognizes this, they will be removed and can later be recomputed, i.e. by the aiProcess_GenSmoothNormals flag.
The step will also remove meshes that are infinitely small and reduce animation tracks consisting of hundreds if redundant keys to a single key. The AI_CONFIG_PP_FID_ANIM_ACCURACY config property decides the accuracy of the check for duplicate animation tracks.

int pyassimp.postprocess.aiProcess_FixInfacingNormals = 0x2000

This step tries to determine which meshes have normal vectors that are facing inwards and inverts them.

The algorithm is simple but effective: the bounding box of all vertices + their normals is compared against the volume of the bounding box of all vertices without their normals. This works well for most objects, problems might occur with planar surfaces. However, the step tries to filter such cases. The step inverts all in-facing normals. Generally it is recommended to enable this step, although the result is not always correct.

int pyassimp.postprocess.aiProcess_FixTexturePaths = 0x200000
int pyassimp.postprocess.aiProcess_FlipUVs = 0x800000

This step flips all UV coordinates along the y-axis and adjusts material settings and bitangents accordingly.

Output UV coordinate system:

1 0y|0y ---------- 1x|0y
2 | |
3 | |
4 | |
5 0x|1y ---------- 1x|1y

You'll probably want to consider this flag if you use Direct3D for rendering. The aiProcess_ConvertToLeftHanded flag supersedes this setting and bundles all conversions typically required for D3D-based applications.

int pyassimp.postprocess.aiProcess_FlipWindingOrder = 0x1000000

This step adjusts the output face winding order to be CW.

The default face winding order is counter clockwise (CCW).

Output face order:

1  x2
2 
3  x0
4 x1

int pyassimp.postprocess.aiProcess_GenEntityMeshes = 0x100000
int pyassimp.postprocess.aiProcess_GenNormals = 0x20

Generates normals for all faces of all meshes.

This is ignored if normals are already there at the time this flag is evaluated. Model importers try to load them from the source file, so they're usually already there. Face normals are shared between all points of a single face, so a single point can have multiple normals, which forces the library to duplicate vertices in some cases. aiProcess_JoinIdenticalVertices is #senseless# then.

This flag may not be specified together with aiProcess_GenSmoothNormals.

int pyassimp.postprocess.aiProcess_GenSmoothNormals = 0x40

Generates smooth normals for all vertices in the mesh.

This is ignored if normals are already there at the time this flag is evaluated. Model importers try to load them from the source file, so they're usually already there.

This flag may not be specified together with aiProcess_GenNormals. There's a configuration option, AI_CONFIG_PP_GSN_MAX_SMOOTHING_ANGLE<tt> which allows you to specify an angle maximum for the normal smoothing algorithm. Normals exceeding this limit are not smoothed, resulting in a 'hard' seam between two faces. Using a decent angle here (e.g. 80 degrees) results in very good visual appearance.

int pyassimp.postprocess.aiProcess_GenUVCoords = 0x40000

This step converts non-UV mappings (such as spherical or cylindrical mapping) to proper texture coordinate channels.

Most applications will support UV mapping only, so you will probably want to specify this step in every case. Note that Assimp is not always able to match the original mapping implementation of the 3D app which produced a model perfectly. It's always better to let the modelling app compute the UV channels - 3ds max, Maya, Blender, LightWave, and Modo do this for example.

Note
If this step is not requested, you'll need to process the AI_MATKEY_MAPPING<tt> material property in order to display all assets properly.
int pyassimp.postprocess.aiProcess_ImproveCacheLocality = 0x800

Reorders triangles for better vertex cache locality.

The step tries to improve the ACMR (average post-transform vertex cache miss ratio) for all meshes. The implementation runs in O(n) and is roughly based on the 'tipsify' algorithm (see this paper). If you intend to render huge models in hardware, this step might be of interest to you. The #AI_CONFIG_PP_ICL_PTCACHE_SIZE<tt>config setting can be used to fine-tune the cache optimization.

int pyassimp.postprocess.aiProcess_JoinIdenticalVertices = 0x2

Identifies and joins identical vertex data sets within all imported meshes.

After this step is run, each mesh contains unique vertices, so a vertex may be used by multiple faces. You usually want to use this post processing step. If your application deals with indexed geometry, this step is compulsory or you'll just waste rendering time. If this flag is not specified, no vertices are referenced by more than one face and no index buffer is required for rendering.

int pyassimp.postprocess.aiProcess_LimitBoneWeights = 0x200

Limits the number of bones simultaneously affecting a single vertex to a maximum value.

If any vertex is affected by more than the maximum number of bones, the least important vertex weights are removed and the remaining vertex weights are renormalized so that the weights still sum up to 1. The default bone weight limit is 4 (defined as AI_LMW_MAX_WEIGHTS<tt> in config.h), but you can use the AI_CONFIG_PP_LBW_MAX_WEIGHTS<tt> setting to supply your own limit to the post processing step.

If you intend to perform the skinning in hardware, this post processing step might be of interest to you.

int pyassimp.postprocess.aiProcess_MakeLeftHanded = 0x4

Converts all the imported data to a left-handed coordinate space.

By default the data is returned in a right-handed coordinate space (which OpenGL prefers). In this space, +X points to the right, +Z points towards the viewer, and +Y points upwards. In the DirectX coordinate space +X points to the right, +Y points upwards, and +Z points away from the viewer.

You'll probably want to consider this flag if you use Direct3D for rendering. The aiProcess_ConvertToLeftHanded flag supersedes this setting and bundles all conversions typically required for D3D-based applications.

int pyassimp.postprocess.aiProcess_OptimizeAnimations = 0x200000
int pyassimp.postprocess.aiProcess_OptimizeGraph = 0x400000

A postprocessing step to optimize the scene hierarchy.

Nodes without animations, bones, lights or cameras assigned are collapsed and joined.

Node names can be lost during this step. If you use special 'tag nodes' to pass additional information through your content pipeline, use the AI_CONFIG_PP_OG_EXCLUDE_LIST<tt> setting to specify a list of node names you want to be kept. Nodes matching one of the names in this list won't be touched or modified.

Use this flag with caution. Most simple files will be collapsed to a single node, so complex hierarchies are usually completely lost. This is not useful for editor environments, but probably a very effective optimization if you just want to get the model data, convert it to your own format, and render it as fast as possible.

This flag is designed to be used with aiProcess_OptimizeMeshes for best results.

Note
'Crappy' scenes with thousands of extremely small meshes packed in deeply nested nodes exist for almost all file formats. aiProcess_OptimizeMeshes in combination with aiProcess_OptimizeGraph usually fixes them all and makes them renderable.

int pyassimp.postprocess.aiProcess_OptimizeMeshes = 0x200000

A postprocessing step to reduce the number of meshes.

This will, in fact, reduce the number of draw calls.

This is a very effective optimization and is recommended to be used together with aiProcess_OptimizeGraph, if possible. The flag is fully compatible with both aiProcess_SplitLargeMeshes and aiProcess_SortByPType.

int pyassimp.postprocess.aiProcess_PreTransformVertices = 0x100

Removes the node graph and pre-transforms all vertices with the local transformation matrices of their nodes.

The output scene still contains nodes, however there is only a root node with children, each one referencing only one mesh, and each mesh referencing one material. For rendering, you can simply render all meshes in order - you don't need to pay attention to local transformations and the node hierarchy. Animations are removed during this step. This step is intended for applications without a scenegraph. The step CAN cause some problems: if e.g. a mesh of the asset contains normals and another, using the same material index, does not, they will be brought together, but the first meshes's part of the normal list is zeroed. However, these artifacts are rare.

Note
The AI_CONFIG_PP_PTV_NORMALIZE<tt> configuration property can be set to normalize the scene's spatial dimension to the -1...1 range.
int pyassimp.postprocess.aiProcess_RemoveComponent = 0x10

Removes some parts of the data structure (animations, materials, light sources, cameras, textures, vertex components).

The components to be removed are specified in a separate configuration option, AI_CONFIG_PP_RVC_FLAGS<tt>. This is quite useful if you don't need all parts of the output structure. Vertex colors are rarely used today for example... Calling this step to remove unneeded data from the pipeline as early as possible results in increased performance and a more optimized output data structure. This step is also useful if you want to force Assimp to recompute normals or tangents. The corresponding steps don't recompute them if they're already there (loaded from the source asset). By using this step you can make sure they are NOT there.

This flag is a poor one, mainly because its purpose is usually misunderstood. Consider the following case: a 3D model has been exported from a CAD app, and it has per-face vertex colors. Vertex positions can't be shared, thus the aiProcess_JoinIdenticalVertices step fails to optimize the data because of these nasty little vertex colors. Most apps don't even process them, so it's all for nothing. By using this step, unneeded components are excluded as early as possible thus opening more room for internal optimizations.

int pyassimp.postprocess.aiProcess_RemoveRedundantMaterials = 0x1000

Searches for redundantunreferenced materials and removes them.

This is especially useful in combination with the #aiProcess_PretransformVertices and aiProcess_OptimizeMeshes flags. Both join small meshes with equal characteristics, but they can't do their work if two meshes have different materials. Because several material settings are lost during Assimp's import filters, (and because many exporters don't check for redundant materials), huge models often have materials which are are defined several times with exactly the same settings.

Several material settings not contributing to the final appearance of a surface are ignored in all comparisons (e.g. the material name). So, if you're passing additional information through the content pipeline (probably using #magic# material names), don't specify this flag. Alternatively take a look at the AI_CONFIG_PP_RRM_EXCLUDE_LIST<tt> setting.

int pyassimp.postprocess.aiProcess_SortByPType = 0x8000

This step splits meshes with more than one primitive type in homogeneous sub-meshes.

The step is executed after the triangulation step. After the step returns, just one bit is set in aiMesh::mPrimitiveTypes. This is especially useful for real-time rendering where point and line primitives are often ignored or rendered separately. You can use the AI_CONFIG_PP_SBP_REMOVE<tt> option to specify which primitive types you need. This can be used to easily exclude lines and points, which are rarely used, from the import.

int pyassimp.postprocess.aiProcess_SplitByBoneCount = 0x2000000

This step splits meshes with many bones into sub-meshes so that each su-bmesh has fewer or as many bones as a given limit.

int pyassimp.postprocess.aiProcess_SplitLargeMeshes = 0x80

Splits large meshes into smaller sub-meshes.

This is quite useful for real-time rendering, where the number of triangles which can be maximally processed in a single draw-call is limited by the video driverhardware. The maximum vertex buffer is usually limited too. Both requirements can be met with this step: you may specify both a triangle and vertex limit for a single mesh.

The split limits can (and should!) be set through the AI_CONFIG_PP_SLM_VERTEX_LIMIT<tt> and AI_CONFIG_PP_SLM_TRIANGLE_LIMIT<tt> settings. The default values are AI_SLM_DEFAULT_MAX_VERTICES<tt> and AI_SLM_DEFAULT_MAX_TRIANGLES<tt>.

Note that splitting is generally a time-consuming task, but only if there's something to split. The use of this step is recommended for most users.

int pyassimp.postprocess.aiProcess_TransformUVCoords = 0x80000

This step applies per-texture UV transformations and bakes them into stand-alone vtexture coordinate channels.

UV transformations are specified per-texture - see the AI_MATKEY_UVTRANSFORM<tt> material key for more information. This step processes all textures with transformed input UV coordinates and generates a new (pre-transformed) UV channel which replaces the old channel. Most applications won't support UV transformations, so you will probably want to specify this step.

Note
UV transformations are usually implemented in real-time apps by transforming texture coordinates at vertex shader stage with a 3x3 (homogenous) transformation matrix.

int pyassimp.postprocess.aiProcess_Triangulate = 0x8

Triangulates all faces of all meshes.

By default the imported mesh data might contain faces with more than 3 indices. For rendering you'll usually want all faces to be triangles. This post processing step splits up faces with more than 3 indices into triangles. Line and point primitives are #not# modified! If you want 'triangles only' with no other kinds of primitives, try the following solution:

int pyassimp.postprocess.aiProcess_ValidateDataStructure = 0x400

Validates the imported scene data structure. This makes sure that all indices are valid, all animations and bones are linked correctly, all material references are correct .. etc.

It is recommended that you capture Assimp's log output if you use this flag, so you can easily find out what's wrong if a file fails the validation. The validator is quite strict and will find #all# inconsistencies in the data structure... It is recommended that plugin developers use it to debug their loaders. There are two types of validation failures:

  • Error: There's something wrong with the imported data. Further postprocessing is not possible and the data is not usable at all. The import fails. #Importer::GetErrorString() or aiGetErrorString() carry the error message around.
  • Warning: There are some minor issues (e.g. 1000000 animation keyframes with the same time), but further postprocessing and use of the data structure is still safe. Warning details are written to the log file, AI_SCENE_FLAGS_VALIDATION_WARNING<tt> is set in aiScene::mFlags<li>
    post-processing step is not time-consuming. Its use is not compulsory, but recommended.
tuple pyassimp.postprocess.aiProcessPreset_TargetRealtime_Fast
Initial value:
1 = ( \
2  aiProcess_CalcTangentSpace | \
3  aiProcess_GenNormals | \
4  aiProcess_JoinIdenticalVertices | \
5  aiProcess_Triangulate | \
6  aiProcess_GenUVCoords | \
7  aiProcess_SortByPType | \
8  0 )
tuple pyassimp.postprocess.aiProcessPreset_TargetRealtime_MaxQuality
Initial value:
1 = ( \
2  aiProcessPreset_TargetRealtime_Quality | \
3  aiProcess_FindInstances | \
4  aiProcess_ValidateDataStructure | \
5  aiProcess_OptimizeMeshes | \
6  0 )
tuple pyassimp.postprocess.aiProcessPreset_TargetRealtime_Quality
Initial value:
1 = ( \
2  aiProcess_CalcTangentSpace | \
3  aiProcess_GenSmoothNormals | \
4  aiProcess_JoinIdenticalVertices | \
5  aiProcess_ImproveCacheLocality | \
6  aiProcess_LimitBoneWeights | \
7  aiProcess_RemoveRedundantMaterials | \
8  aiProcess_SplitLargeMeshes | \
9  aiProcess_Triangulate | \
10  aiProcess_GenUVCoords | \
11  aiProcess_SortByPType | \
12  aiProcess_FindDegenerates | \
13  aiProcess_FindInvalidData | \
14  0 )