Home / User guide / Step 2: pre-processing

Step 2: pre-processing

Check the tracking quality, keep the well-tracked body points, and prepare the data for clustering.

What happens#

  1. Each keypoint file is converted to a standard format, so output from different DLC versions works.
  2. CUBE decides which body points to keep. A body point is kept when its confidence is high enough in at least 85% of sessions (defaults: minimum confidence 0.30, metric median, at least 6 body points kept). The result is written to bodypart_confidence_report.csv.
  3. When Step 3 loads the data, frames with low likelihood (below 0.30) are filled in from their neighbours, but only across gaps of up to 0.5 s. Longer dropouts are held at the last good value instead of being bridged by a straight line.
  4. A short moving average (70 ms) reduces single-frame jitter without shifting the timing.

The settings are in DLC & Prep Settings (body-point selection) and Advanced CUBE Analysis Parameters → Feature extraction (likelihood threshold, boxcar smoothing).