Heavy Tracking

Heavy Tracking is the primary high-quality workflow for camera reconstruction and solved lens data. Use it when the shot needs strong solve quality, a specific distortion model, calibration/STMap export, masks, undistortion, dense point clouds, or mesh generation.

How to use it

  1. 1 Prepare footage in the Select Footage or Batch Processing panel.
  2. 2 Choose a preset in the Heavy Tracker panel; Medium (Recommended) suits most shots, and the panel suggests one for your footage.
  3. 3 Turn on the options the shot needs: Zoom for zoom shots, Heavy Rotation for strong roll, Mask for moving objects, and Undistort plus Dense for a dense point cloud.
  4. 4 Optionally open Lens Distortion / STMap and choose a camera model; Simple Radial is the default.
  5. 5 Click Autotrack (Heavy) and follow the Progress panel. Max users can cancel a running job.
  6. 6 Check Last Result (registered frames, reprojection error, stutter, point count), then import the result.

Heavy Tracker Panel

The Heavy Tracker is the main, fully automatic camera solver in MotionMaster 3D. It combines COLMAP, a custom CUDA build of GLOMAP and OpenMVS: it extracts and matches features across the whole sequence, verifies the matches geometrically, solves the camera and the lens, and can go on to dense point clouds and meshes. For most shots you only pick a preset and a few toggles.

Preset

Start with the Preset dropdown. Below it, the panel recommends a preset for the selected footage based on its resolution and number of frames. The question mark button opens a settings reference with a footage guide inside Blender.

  • •
    Super Fast: A quick sanity check: a fast first pass to see whether the clip reconstructs at all.
  • •
    Fast: A benchmarked fast preset that is reliable on long footage.
  • •
    Medium (Recommended): The default. A benchmarked balance of speed and quality that suits most shots.
  • •
    High Quality: A benchmarked high-quality preset for maximum reliability.
  • •
    Extra High: Maximum settings, for when every other preset fails.

If a track fails, move to a higher preset before changing individual parameters.

Options

The grid of toggle buttons under the preset turns individual options on and off. Blue means enabled.

  • •
    Zoom (Tracking): Enable when the focal length changes during the shot, such as an optical zoom or a dolly zoom. Frames may then use their own camera calibration.
  • •
    Heavy Rotation (Tracking): Enable for footage with strong roll or tilt, such as FPV drone shots or Dutch angles.
  • •
    Exhaustive (Tracking): Matches every frame pair instead of only neighboring frames. Much slower, but useful for large viewpoint changes or clips with few frames.
  • •
    Cuda (GPU) (Processing): Runs feature extraction and matching on an NVIDIA GPU, which is much faster.
  • •
    Undistort (Processing): Removes lens distortion from the images. Required for Dense.
  • •
    Dense (Processing): Generates a dense, colored point cloud of the scene. Needs Undistort and takes longer, and is required for mesh generation.
  • •
    Accurate (Quality): Affine-shape SIFT descriptors that hold up better on challenging imagery. Forces CPU feature extraction even with Cuda enabled.
  • •
    Robust (Quality): Advanced feature detection with more discriminative descriptors. Also forces CPU feature extraction.
  • •
    Mask (Quality): Limits tracking to parts of the image, for example to exclude moving people, sky or reflections. Shows Track Color and Mask Folder.

When Accurate or Robust is on together with Cuda, the panel warns that CPU SIFT is used and the GPU is disabled for feature extraction, so expect a slower run. With Mask enabled, choose the Track Color (Use Black tracks the black, unmasked background; Use White tracks the white, masked region) and the Mask Folder, or create the masks in the Mask panel.

Lens Distortion / STMap

This section chooses the lens model COLMAP solves (Max). Camera Model defaults to Simple Radial, and the line below it lists the parameters of the chosen model. Leave it on Auto to let COLMAP estimate them, or tick Use Known Camera Params and enter every value in the order shown. Twelve models are available, from Simple Pinhole to fisheye and thin-prism models.

After a solve, Solved Result picks which Heavy run the lens data comes from. Export Calibration writes the full camera data to a JSON file, and Export STMaps writes 32-bit undistort and redistort maps for Nuke, Fusion and other compositors. The complete workflow is on Lens calibration and STMaps.

Advanced Parameters

This section shows every value the presets control. Changing the preset updates them all, and you can override any value by hand afterwards.

Processing

  • •
    Overlap: how many neighboring frames each frame is matched with. Higher values are more robust but slower.
  • •
    Features: the maximum number of features extracted per image. More features give more detail at the cost of time.
  • •
    Max Size: the largest image size used for processing. Bigger images are downscaled; lower it if you run out of memory.

Sequential Matcher

  • •
    Max Matches: The maximum number of matches per image pair. Limits memory use and processing time.
  • •
    Confidence: The threshold for geometric verification. Higher values are stricter and filter out more bad matches, but can also discard good ones.
  • •
    Max Trials: How many attempts are made to fit the geometric model. More trials are more reliable but slower.
  • •
    Max Ratio: How distinctive a match must be compared with the next best one. Lower values keep fewer but more reliable matches.
  • •
    Max Error: The largest pixel error a match may have during verification. Lower values are stricter.
  • •
    Min Inlier Ratio: The minimum share of matches that must agree for an image pair to be accepted.
  • •
    Min Inliers: The minimum number of agreeing matches needed to link two images.

Loop Closure

Enable Loop Detection recognizes when the camera returns to an area it has already seen, such as when walking in a circle or orbiting an object, and ties those frames together. When enabled, extra fields control how often and how thoroughly loops are searched for.

Track Establishment

  • •
    Tracks/View: the minimum number of tracked points an image needs to be included in the reconstruction.
  • •
    Views/Track: the minimum number of images that must see a 3D point for it to be kept.

Optimization

  • •
    Global Pos: the threshold used during global positioning of the cameras.
  • •
    Bundle Adj: the threshold used during bundle adjustment, the final joint refinement of cameras and points.
  • •
    Tri Matches: the minimum number of matches needed to triangulate a new 3D point. Lower values give more but noisier points.

Autotrack (Heavy)

Click Autotrack (Heavy) to start. The Progress panel at the top shows the current stage, and Max users can cancel a running job there and keep the partial output. With Batch Mode on, the button points you to the Batch Processing panel instead.

Last Result

After a run, the panel summarizes the solve so you can judge it before importing:

  • •
    Reg: the share of frames that were registered (solved). 100% means every frame has a camera.
  • •
    Reproj: the mean reprojection error in pixels. Lower is better; around 1 px or below is good for most VFX work.
  • •
    Stutter: how jerky the solved camera path is, from 0 to 1. Lower means a smoother camera.
  • •
    Pts: the number of reconstructed 3D points.

Mesh Generation

When a Heavy result has a dense point cloud (Undistort and Dense enabled), you can build a mesh from it. Choose the Quality: Low (about 100K faces), Medium (about 500K), High (about 1.5M), Ultra (about 5M) or Custom, which exposes the dense resolution, target face count, refinement and texture settings. Max Threads limits CPU threads (0 uses all of them). Click Generate Mesh to start.

The mesh can only be as good as the dense cloud. If the dense point cloud looks noisy, improve the footage, masks or preset and track again before generating a mesh.

Technical notes

Heavy Tracking uses the bundled COLMAP/GLOMAP toolchain for feature extraction, matching, mapping, camera reconstruction, and optional image undistortion, then uses OpenMVS-related tools for dense and mesh stages.

After a solve, the Solved Result dropdown can target any Heavy result found in the active Import Results folder. Calibration and STMap exports always use that selected result instead of silently using the most recent run.

Docs Lens Calibration and STMap Export

The Heavy Tracking lens workflow preserves the complete solved COLMAP camera model and exports portable calibration JSON plus float STMaps for undistorting the original plate and redistorting CG back into it.

Docs AI Masks and Polygon Masks

The masking system helps Heavy Tracking ignore or isolate moving objects. It supports SAM2 AI mask layers, manually drawn polygon layers, keyframed polygon animation, and composed mask output.

Docs Dense Point Clouds

Dense Point Cloud generation creates a high-detail colorized representation of the tracked scene from Heavy Tracking results. It is useful when sparse points are not enough to understand the environment.

 Tutorial What Makes Footage Good for Camera Tracking? 9 Factors With Real Examples Tutorial How to Shoot Footage for Camera Tracking: Settings, Moves and an On-Set Checklist Tutorial How to Camera Track iPhone Footage in Blender, Step by Step Guide Blender camera tracking addon

A complete camera tracking addon built for Blender VFX artists

 Guide 3D reconstruction in Blender

Turn tracked footage into dense point clouds and mesh reference in Blender