Decimation & LOD
Introduction​
Dense meshes from photogrammetry scans or detailed Sketchfab downloads can overwhelm Blender's viewport — navigation slows, modifiers bog down, and the scene becomes difficult to work with. Decimation is the process of reducing polygon count while preserving as much of the original shape as possible.
This page covers Blender's Decimate modifier and manual decimation strategies, along with a practical approach to managing multiple levels of detail (LOD) for the same object within a scene.
Polygon Reduction
Reduce vertex count by 50-95% while maintaining recognizable form.
Viewport Performance
Keep scenes responsive by using appropriate detail levels for each object.
Level of Detail
Manage multiple resolutions of the same model for different purposes.
The Decimate Modifier​
Blender's Decimate modifier provides three reduction modes, each suited to different situations.
Collapse​
The default mode. Iteratively merges the edge that causes the least visible change until the target ratio is reached.
- Ratio: 0.0 to 1.0 (1.0 = no reduction, 0.1 = 90% reduction)
- Best for: Organic forms, photogrammetry output, smooth surfaces
- Limitation: Can produce triangle-heavy output with unpredictable topology
Un-Subdivide​
Reverses subdivision by removing every other edge loop. Only works well on meshes that were created by subdivision.
- Iterations: Number of subdivision levels to reverse
- Best for: Meshes that were subdivided evenly (sculpted models, SubD output)
- Limitation: Produces poor results on irregular topology
Planar​
Dissolves faces that are nearly coplanar — faces that share a similar angle. Useful for architectural or CAD-derived geometry where large flat surfaces have been over-tessellated.
- Angle Limit: Faces within this angle of each other are merged
- Best for: Architectural models, CAD imports, flat-surface geometry
- Limitation: Rounds off edges and corners if the angle limit is too high
Practical Workflow​
Step 1: Duplicate Before Decimating​
Never decimate your only copy. Always work on a duplicate:
- Select the object
- Duplicate (Shift + D) and move to a separate collection (e.g.,
LOD_Low) - Keep the original in a
LOD_HighorSourcecollection, hidden but preserved
Step 2: Choose the Right Mode​
| Mesh Type | Recommended Mode | Starting Ratio/Setting |
|---|---|---|
| Photogrammetry scan | Collapse | 0.1 – 0.3 |
| Subdivided model | Un-Subdivide | 1-2 iterations |
| CAD/architectural | Planar | 5° – 15° angle limit |
| General-purpose | Collapse | 0.5 then adjust |
Step 3: Evaluate the Result​
Before applying the modifier:
- Rotate around the object — check for visual artifacts, collapsed features, or sharp edges that became rounded
- Compare the silhouette against the original (toggle visibility of the source collection)
- Check the face count in the modifier panel — it shows the resulting polygon count in real-time
Step 4: Apply and Clean Up​
Once satisfied:
- Apply the modifier (Ctrl + A on the modifier panel)
- Run Mesh Repair checks — decimation can introduce non-manifold edges or flipped normals
- Use Merge by Distance to clean up any near-coincident vertices the algorithm left behind
Managing Levels of Detail​
For investigative scenes with multiple objects, maintaining a single detail level for everything is impractical. A scene might contain:
- A high-detail vehicle (the subject of analysis)
- Medium-detail surrounding buildings
- Low-detail background objects and terrain
LOD Strategy with Collections​
Organize detail levels using Blender's collection system:
03_Evidence/
├── Vehicle_High (127,000 faces — for analysis and close-up renders)
├── Vehicle_Low (8,000 faces — for viewport navigation and scene assembly)
Toggle visibility between collections depending on the task:
- Scene assembly / camera animation: Use low-detail versions
- Close-up analysis / final render: Switch to high-detail
Viewport Simplification​
Beyond per-object decimation, Blender offers global viewport optimization:
- Simplify panel (Render Properties → Simplify): Set a max subdivision level for the viewport
- Object bounding box display: In the Outliner, set objects to display as Bounds for maximum viewport performance
- Local View (/ on numpad): Isolate individual objects to work on them without the scene overhead
Common Pitfalls​
- Decimating the only copy: Always duplicate first. You cannot recover detail after applying a Decimate modifier.
- Over-decimation: Aggressive ratios (< 0.05) can destroy the shape entirely. Decimate in stages — go from 1.0 to 0.5, evaluate, then from 0.5 to 0.25.
- Ignoring the mode: Using Collapse on a flat architectural surface produces messy triangulation. Planar mode dissolves the same geometry cleanly.
- Skipping cleanup: Decimation can create non-manifold edges and flipped normals. Always run a repair pass afterward.
Summary​
Key Takeaways:
- Duplicate before decimating — never reduce your only copy.
- Choose the right mode: Collapse for organic, Planar for architectural, Un-Subdivide for evenly subdivided meshes.
- Manage scene performance with LOD collections — keep high and low versions accessible.
- Run mesh repair checks after applying decimation.
Documentation for Methodology​
What to document when using Decimation
In Your Method Section​
Describe the decimation approach used, the target reduction, and the reason for reducing the mesh (e.g., viewport performance, scene assembly, file size).
In Your Decision Log​
Record the following when decimating:
- Original face/vertex count
- Decimate mode and settings (ratio, angle limit, or iterations)
- Final face/vertex count after application
- Visual quality assessment (acceptable silhouette preservation: yes/no)
- Whether the decimated version or original is used for the final analysis
- Date performed
Verification​
Compare the decimated mesh silhouette against the original from multiple angles. Verify that key features (edges, openings, proportional relationships) are preserved. Check for non-manifold geometry introduced by the reduction.
Common Limitations​
Decimation is lossy — detail cannot be recovered once the modifier is applied. The algorithm makes no distinction between important features and noise; both are reduced equally. Fine details like text, small holes, or thin protrusions are lost first.
Example Methodology Statement​
"The photogrammetry-derived vehicle model (487,000 faces) was decimated for viewport performance using Blender's Decimate modifier (Collapse mode, ratio 0.15). The resulting mesh (73,050 faces) preserved the vehicle silhouette and key proportional features. The original high-resolution model was retained in a separate collection for reference. Visual comparison confirmed acceptable shape preservation for scene-assembly use. The high-resolution version was used for all dimensional measurements."