Skip to main content

​
Overview

When working with BIM models in Speckle, you’ll encounter objects whose geometry is represented through displayValue properties. This guide shows you how to extract this geometry, including how to handle instance definitions that require resolving applicationId references. What you’ll learn:
  • How to extract meshes from displayValue properties
  • How to build an applicationId index for reference lookups
  • How to resolve InstanceDefinition references to get actual geometry
  • How to apply transformations to instances
For a comprehensive explanation of proxies and why Speckle uses them, see BIM Data Patterns: Pattern 5 and Proxification.

​
The Challenge with Display Values

After receiving a model version with operations.receive(), objects have a displayValue property containing visualization geometry. This can be:
  1. Direct meshes - The geometry is right there
  2. Instance references - References to InstanceDefinition objects by applicationId
The second case requires building an applicationId index to resolve the references. 
# Case 1: Direct mesh
obj.displayValue = [mesh1, mesh2, ...]  # Meshes are directly accessible

# Case 2: Instance references
obj.displayValue = [
    {
        "applicationId": "instance_def_123",
        # ... other instance properties
        # Need to look this up in instanceDefinitionProxies
    }
]

# InstanceDefinition itself has no 'value' property
# It has 'objects' which are applicationIds that need resolution

​
Extracting Display Values

​
Finding Objects with Display Values

Start by finding all objects that have display geometry: 
from specklepy.api import operations
from specklepy.transports.server import ServerTransport
from specklepy.api.client import SpeckleClient
from specklepy.api.credentials import get_default_account
from specklepy.objects.graph_traversal.traversal import GraphTraversal

# Receive model version
client = SpeckleClient(host="https://app.speckle.systems")
client.authenticate_with_account(get_default_account())
transport = ServerTransport("project_id", client)
version = operations.receive("version_id", transport)

def find_displayable_objects(root):
    """Find all objects with displayValue property."""
    displayable = []
    traversal = GraphTraversal([])

    for context in traversal.traverse(root):
        obj = context.current
        if hasattr(obj, "displayValue") and obj.displayValue:
            displayable.append(obj)

    return displayable

objects_with_display = find_displayable_objects(version)
print(f"Found {len(objects_with_display)} displayable objects")

​
Extracting Direct Meshes

For objects with direct mesh geometry: 
from specklepy.objects.geometry import Mesh

def extract_direct_meshes(obj):
    """
    Extract meshes that are directly in displayValue.

    Returns:
        List of Mesh objects
    """
    if not hasattr(obj, "displayValue"):
        return []

    display = obj.displayValue
    meshes = []

    # Normalize to list
    display_list = [display] if not isinstance(display, list) else display

    for item in display_list:
        # Check if it's a Mesh object
        if isinstance(item, Mesh):
            meshes.append(item)

    return meshes

# Usage
for obj in objects_with_display:
    meshes = extract_direct_meshes(obj)
    for mesh in meshes:
        if hasattr(mesh, "vertices"):
            print(f"Mesh: {len(mesh.vertices) // 3} vertices")

​
Building an applicationId Index

To resolve instance references, build an index mapping applicationIds to objects: 
from specklepy.objects.graph_traversal.traversal import GraphTraversal

def build_applicationid_index(root):
    """
    Build a dictionary mapping applicationId to objects.

    Why this is needed: InstanceDefinition objects reference other
    objects by applicationId, not by direct embedding.

    Returns:
        Dictionary: {applicationId: object}
    """
    index = {}
    traversal = GraphTraversal([])

    for context in traversal.traverse(root):
        obj = context.current

        # Index any object with an applicationId
        if hasattr(obj, "applicationId") and obj.applicationId:
            index[obj.applicationId] = obj

    return index

# Build the index once
app_id_index = build_applicationid_index(version)
print(f"Indexed {len(app_id_index)} objects")

# Now you can quickly look up any object
some_obj = app_id_index.get("some-application-id")
if some_obj:
    print(f"Found: {getattr(some_obj, 'name', 'unnamed')}")
Why build an index? Without it, you’d traverse the entire tree for each lookup. With an index, lookups are instant (O(1) vs O(n)).

​
Working with Instance Definitions

​
Understanding Instance Definitions

Instance definitions enable geometry reuse. The definition contains references to geometry objects (by applicationId), and each instance includes a transformation matrix. 
# Root has instanceDefinitionProxies
version.instanceDefinitionProxies = [
    {
        "id": "door_def_123",
        "name": "Door Type A",
        "objects": [
            "mesh_app_id_1",
            "mesh_app_id_2"
        ]
        # NOTE: No 'value' property like LevelProxy or ColorProxy!
        # 'objects' are applicationIds that need to be resolved
    }
]

# Objects reference the definition
obj.displayValue = [
    {
        "applicationId": "door_def_123",  # References the definition
        "transform": [...]                 # 4x4 transformation matrix
    }
]

​
Extracting Instance Definitions

Extract instance definitions and resolve their geometry references: 
def extract_instance_definitions(root, app_id_index):
    """
    Extract instance definitions and resolve their geometry.

    Args:
        root: Model version object
        app_id_index: applicationId index from build_applicationid_index()

    Returns:
        Dictionary: {applicationId: definition_info}
    """
    if not hasattr(root, "instanceDefinitionProxies"):
        return {}

    definitions = {}

    for proxy in root.instanceDefinitionProxies:
        # Index by applicationId (not id) - instances reference by applicationId
        def_app_id = getattr(proxy, "applicationId", None)

        if def_app_id:
            # Resolve geometry references using the index
            geometry = []
            if hasattr(proxy, "objects"):
                for app_id in proxy.objects:
                    obj = app_id_index.get(app_id)
                    if obj:
                        geometry.append(obj)

            definitions[def_app_id] = {
                "name": getattr(proxy, "name", "unnamed"),
                "geometry": geometry,
                "proxy": proxy
            }

    return definitions

# Usage
definitions = extract_instance_definitions(version_root_object, app_id_index)
for def_app_id, def_info in definitions.items():
    print(f"{def_info['name']}: {len(def_info['geometry'])} objects")

# Args:
#     root: Version object
#     app_id_index: applicationId index
#     definitions: Instance definitions from extract_instance_definitions()

# Returns:
#     List of dictionaries with mesh and metadata

​
Resolving Instance References in Display Values

Now you can resolve instance references when extracting display values: 
from specklepy.objects.geometry import Mesh

def extract_all_display_meshes(root, app_id_index, definitions):
    """
    Extract all meshes from display values, resolving instance references.

    Args:
        root: Version object
        app_id_index: applicationId index
        definitions: Instance definitions from extract_instance_definitions()

    Returns:
        List of dictionaries with mesh and metadata
    """
    results = []
    traversal = GraphTraversal([])

    for context in traversal.traverse(root):
        obj = context.current

        if not hasattr(obj, "displayValue"):
            continue

        display = obj.displayValue
        display_list = [display] if not isinstance(display, list) else display

        for item in display_list:
            # Case 1: Direct mesh
            if isinstance(item, Mesh):
                results.append({
                    "mesh": item,
                    "source": obj,
                    "is_instance": False
                })

            # Case 2: Instance reference (by definitionId which is applicationId)
            elif hasattr(item, "definitionId"):
                def_app_id = item.definitionId
                if def_app_id in definitions:
                    # Get the geometry from the definition
                    def_info = definitions[def_app_id]
                    for geom in def_info["geometry"]:
                        if isinstance(geom, Mesh):
                            results.append({
                                "mesh": geom,
                                "source": obj,
                                "is_instance": True,
                                "definition": def_info,
                                "transform": getattr(item, "transform", None)
                            })

    return results

​
Applying Transformations

Instance references include transformation matrices that position, rotate, and scale the geometry: 
def get_transformed_instances(root, app_id_index):
    """
    Extract instance meshes with their transformations.

    Note: This function identifies which meshes need transformation.
    Actual transformation math is shown for reference but may require
    additional libraries (numpy) for production use.

    Returns:
        List of dictionaries with mesh and transform info
    """
    definitions = extract_instance_definitions(root, app_id_index)
    instances = []
    traversal = GraphTraversal([])

    for context in traversal.traverse(root):
        obj = context.current

        if not hasattr(obj, "displayValue"):
            continue

        display = obj.displayValue
        display_list = [display] if not isinstance(display, list) else display

        for item in display_list:
            # Look for instance references with transforms
            if hasattr(item, "applicationId") and hasattr(item, "transform"):
                def_id = item.applicationId
                transform = item.transform

                if def_id in definitions:
                    def_info = definitions[def_id]

                    for geom in def_info["geometry"]:
                        instances.append({
                            "geometry": geom,
                            "definition_name": def_info["name"],
                            "transform": transform,  # 4x4 matrix as flat list
                            "source_object": obj
                        })

    return instances

# Usage
instances = get_transformed_instances(version, app_id_index)
print(f"Found {len(instances)} instance uses")
Transform structure: The transform is a 4x4 transformation matrix stored as a flat list of 16 numbers (column-major order). This represents position, rotation, and scale.

​
Complete Example

Here’s a complete workflow combining all concepts: 
from specklepy.api import operations
from specklepy.transports.server import ServerTransport
from specklepy.api.client import SpeckleClient
from specklepy.api.credentials import get_default_account
from specklepy.objects.graph_traversal.traversal import GraphTraversal
from specklepy.objects.geometry import Mesh

def analyze_display_geometry(project_id, version_id):
    """Complete analysis of display geometry including instances."""

    # 1. Receive version
    client = SpeckleClient(host="https://app.speckle.systems")
    client.authenticate_with_account(get_default_account())
    transport = ServerTransport(project_id, client)
    version = operations.receive(version_id, transport)

    # 2. Build applicationId index
    print("Building applicationId index...")
    app_id_index = build_applicationid_index(version)
    print(f"Indexed {len(app_id_index)} objects")

    # 3. Extract instance definitions
    print("\nExtracting instance definitions...")
    definitions = extract_instance_definitions(version, app_id_index)
    print(f"Found {len(definitions)} instance definitions:")
    for def_id, def_info in definitions.items():
        print(f"  {def_info['name']}: {len(def_info['geometry'])} objects")

    # 4. Extract all display meshes
    print("\nExtracting display geometry...")
    traversal = GraphTraversal([])

    direct_mesh_count = 0
    instance_mesh_count = 0

    for context in traversal.traverse(version):
        obj = context.current

        if not hasattr(obj, "displayValue"):
            continue

        display = obj.displayValue
        display_list = [display] if not isinstance(display, list) else display

        for item in display_list:
            # Direct mesh
            if isinstance(item, Mesh):
                direct_mesh_count += 1
                vertex_count = len(item.vertices) // 3
                print(f"  Direct mesh: {vertex_count} vertices")

            # Instance reference
            elif hasattr(item, "definitionId"):
                def_app_id = item.definitionId
                if def_app_id in definitions:
                    def_info = definitions[def_app_id]
                    mesh_count = sum(1 for g in def_info["geometry"] if isinstance(g, Mesh))
                    instance_mesh_count += mesh_count

                    has_transform = hasattr(item, "transform")
                    print(f"  Instance of '{def_info['name']}': {mesh_count} meshes"
                          f"{' (with transform)' if has_transform else ''}")

    print(f"\nSummary:")
    print(f"  Direct meshes: {direct_mesh_count}")
    print(f"  Instance meshes: {instance_mesh_count}")
    print(f"  Total: {direct_mesh_count + instance_mesh_count}")

# Run analysis
analyze_display_geometry("your_project_id", "version_id")

​
Best Practices

​
Build the Index Once

Building an applicationId index traverses the entire object tree. Do this once at the start: 
# ✅ Good: Build once, use many times
app_id_index = build_applicationid_index(version)
definitions = extract_instance_definitions(version, app_id_index)

# Then use the index for all lookups
for app_id in some_list_of_ids:
    obj = app_id_index.get(app_id)

# ❌ Bad: Don't rebuild in loops

​
Use Safe Dictionary Access

Always use .get() when looking up by applicationId: 
# ❌ This can crash
obj = app_id_index[app_id]  # KeyError if not found

# ✅ This is safe
obj = app_id_index.get(app_id)
if obj:
    # Process the object
    print(getattr(obj, "name", "unnamed"))

​
Check for Proxy Collections

Not all model versions have instance definitions: 
# ✅ Always check first
if hasattr(version, "instanceDefinitionProxies"):
    definitions = extract_instance_definitions(version, app_id_index)
else:
    print("No instance definitions in this version")
    definitions = {}

# Then safely use the definitions dictionary

​
Troubleshooting

Problem: Instance definition’s objects list has applicationIds but you can’t find the geometry.Why this happens: The geometry objects might not have been indexed, or the applicationIds are incorrect.Solution: Verify your index contains the referenced objects:
def debug_instance_definition(proxy, app_id_index):
    """Debug helper to see what's in an instance definition."""
    print(f"Definition: {getattr(proxy, 'name', 'unnamed')}")
    print(f"  ID: {getattr(proxy, 'id', 'none')}")
    print(f"  References {len(proxy.objects)} objects:")

    for app_id in proxy.objects:
        obj = app_id_index.get(app_id)
        if obj:
            obj_type = getattr(obj, "speckle_type", "unknown")
            print(f"    ✓ {app_id}: {obj_type}")
        else:
            print(f"    ✗ {app_id}: NOT FOUND IN INDEX")

# Use it to debug
for proxy in version.instanceDefinitionProxies:
    debug_instance_definition(proxy, app_id_index)
Problem: Some objects have direct meshes, others have instance references in the same displayValue list.Why this happens: Complex objects can combine directly modeled geometry with instanced components.Solution: Check the type of each item in displayValue:
from specklepy.objects.geometry import Mesh

def extract_mixed_display(obj, definitions):
    """Handle displayValue with both meshes and instances."""
    if not hasattr(obj, "displayValue"):
        return []

    meshes = []
    display_list = [obj.displayValue] if not isinstance(obj.displayValue, list) else obj.displayValue

    for item in display_list:
        # Type 1: Direct mesh
        if isinstance(item, Mesh):
            meshes.append(item)

        # Type 2: Instance reference
        elif hasattr(item, "applicationId"):
            def_id = item.applicationId
            if def_id in definitions:
                # Get meshes from the definition
                for geom in definitions[def_id]["geometry"]:
                    if isinstance(geom, Mesh):
                        meshes.append(geom)

    return meshes
Problem: version.instanceDefinitionProxies doesn’t exist or is empty.Why this happens: Not all models use instancing—depends on the source application and how the model was created.Solution: Always check before accessing, and handle both cases:
# Check if instances are used
if hasattr(version, "instanceDefinitionProxies") and version.instanceDefinitionProxies:
    print(f"Found {len(version.instanceDefinitionProxies)} instance definitions")
    definitions = extract_instance_definitions(version, app_id_index)
else:
    print("No instance definitions - all geometry is direct")
    definitions = {}

# Your code works with or without instances
for obj in objects_with_display:
    # This handles both cases
    meshes = extract_direct_meshes(obj)  # Gets direct meshes
    # If needed, also check for instances using definitions dict
Problem: Can’t tell if displayValue contains meshes or instance references.Why this happens: Both are in the same displayValue property.Solution: Check for the presence of applicationId:
def is_instance_reference(item):
    """Check if displayValue item is an instance reference."""
    # Instance references have applicationId
    # Direct meshes are Mesh objects
    from specklepy.objects.geometry import Mesh

    if isinstance(item, Mesh):
        return False

    if hasattr(item, "applicationId"):
        return True

    return False

# Use it
for item in obj.displayValue:
    if is_instance_reference(item):
        print(f"Instance: {item.applicationId}")
    else:
        print("Direct mesh")

​
Summary

Key concepts:
  • ✅ displayValue can contain direct meshes or instance references
  • ✅ InstanceDefinition objects reference geometry by applicationId, not direct embedding
  • ✅ applicationId index enables fast lookups when resolving references
  • ✅ Transformations position, rotate, and scale instanced geometry
Core workflow:
  1. Build applicationId index once
  2. Extract instance definitions, resolving their geometry references
  3. Process displayValue, handling both direct meshes and instance references
  4. Apply transformations where present
For more on proxies:

​
Next Steps

BIM Data Patterns

Level proxies, color proxies, and organizational hierarchies

Finding and Extracting Data

Advanced traversal and filtering techniques

Understanding Speckle Mesh

Deep dive into mesh structure and encoding

Working with Geometry

Create and manipulate geometry objects
Last modified on January 28, 2026