Nodes

Auto Track

The main solver. Turns an image sequence into a solved camera and a sparse point cloud.

The node that does the actual tracking. It finds features in every frame, matches them across the shot, and solves both the camera motion and a sparse cloud of 3D points.

Inputs

Socket	Required	What it does
`Image Sequence`	Yes	The frames to solve
`Lens Data`	No	Known lens values, from Lens Grid Calibrate or 3DE Lens Import
`Mask Frames`	No	Areas to exclude or restrict to

Output is Auto Track, carrying the cameras and the point cloud together.

Quality

How many features to extract per frame.

Setting	Features	When
Low	2,048	Quick check that the shot solves at all
Medium	3,072	The default. Start here.
High	6,144	When Medium registers only part of the shot

Raising this is the first thing to try when a solve half-completes.

Mask Mode

Appears once something is wired into Mask Frames.

Mode	Meaning
Exclude Masked Area	Ignore what the mask covers. Use this to take a moving subject out of a camera solve.
Limit to Masked Area	Use only what the mask covers. Use this to solve the subject itself.

One mask covers both jobs, so there's no need to invert anything. That's what makes Object Tracking straightforward.

Features

SIFT is the classic algorithm and the default.

LightGlue (AI) uses learned features and matching instead. Much more robust on steep angles, repetitive texture like brickwork, and low-texture surfaces. It needs the AI runtime installed and wants a GPU. Loop detection doesn't run on this backend.

Loop Detection

Spots when the camera returns somewhere it has already been and adds matches between those frames, which pulls out accumulated drift.

Worth turning on for shots that orbit, circle a room, or pan away and come back. Leave it off otherwise, because it indexes every frame at roughly half a second to a second each and gains nothing on a shot that never revisits anything.

Needs SIFT, so it's skipped on LightGlue.

Lens Settings

Calibrate Lens is on by default and refines focal length and distortion while solving. Turn it off when you've fed in lens data you trust and don't want it changed.

Camera Model picks the lens model. Tick Fisheye to see the fisheye options. Only shown while Calibrate Lens is on.

Calculate Zoom allows focal length to vary per frame. Turn it on only when the lens genuinely zooms.

Solver Tuning

The four values at the bottom rarely need touching.

Setting	What it does
Min Tracks/View	Fewest tracks a frame needs to be kept
Min Views/Track	Fewest frames a 3D point must appear in
Loss Threshold	Outlier tolerance during global positioning. Lower is stricter.
BA Loss Threshold	The same during bundle adjustment

Tightening the thresholds cleans up a noisy solve but discards points.

After Solving

The node reports registered cameras, point count and mean error.

Next step	Node
Drop high-error points	Refine Cloud
Reinforce with hand tracks	Enhance Auto Track
Split camera, cloud and lens apart	Separate Track Elements
Build a dense cloud	Densify
Straighten the plate	Lens Distortion
Put it in the scene	Output 3D