DIS Flow (spike)

SPIKE: Dense Inverse Search optical flow, measuring the classical quality gap against Lucas–Kanade.

What it does

DIS Flow estimates motion between two images with Dense Inverse Search: a resolution pyramid, a coarse-to-fine walk, then four warp-and-solve passes at each level. It emits the same four outputs as Optical Flow, a flow map, a colour visualisation, a magnitude field and the frame's peak motion, so anything that consumes one consumes the other.

It is a measurement node. It was built to size the quality gap between Dense Inverse Search and the Lucas-Kanade estimator Optical Flow uses, and the measurement did not favour it.

When to use it

ParameterTypeDefaultWhat it does
imageRasterImage to compute flow from (any Raster source — video, camera, etc.)
referenceRasterExplicit reference frame. If unconnected, emits no motion (spike measures the explicit-pair path).
qualityString"fast"Quality: fast (3 pyramid levels), balanced (5), accurate (6). Higher = larger resolvable motion. One of: fast, balanced, accurate.
enabledBooleantrueEnable optical flow computation
flow_scaleNumber1Multiply flow vectors (amplify or reduce detected motion)
thresholdNumber0Discard flow below this confidence (noise reduction)
positionVec3(0, 0, 0)Position offset in world units (X, Y, Z)
opacityNumber1Layer opacity (0 = transparent, 1 = opaque). Editable, wirable, keyframable.

Gotchas

Reference has no automatic previous frame. Left unwired the node emits a valid, permanently zero result rather than an error. Optical Flow buffers the previous frame; this node measures the explicit-pair path only.

Quality sets the pyramid depth: fast is 3 levels, balanced 5, accurate 6. Deeper resolves larger motion.

Threshold discards flow below a confidence level, not below a magnitude.

Position is declared and never read. Moving it does nothing.

The node ships in the configuration its own measurement condemned. At one refinement pass it is identical to Optical Flow; at the four passes it is fixed at, the shared solve's noise is amplified and the result is measurably worse. The pass count is not a parameter.

Worked example

  1. Wire a video or camera into Image, and the frame to compare against into Reference.

  2. Wire the same pair into Optical Flow.

  3. Raise Quality to accurate on both if the motion is large.

  4. Compare the two Visualization outputs side by side.

See also

Optical Flow · Motion Field · Displace · Rasters (images)