Spend the Labels at the Rim

Allocating a Fixed Annotation Budget for Engulfment-Onset Detection in Macrophage Time-Lapse

A research poster from the School of Emergent Priorities trains a 3D convolutional detector for the onset of engulfment across 412 macrophage time-lapse movies and compares four ways of spending one 2,000-frame labelling budget. Labels drawn within six frames of tracker-detected first contact brought median onset error to 1.1 frames, against 3.1 for uniform sampling and 2.2 for uncertainty sampling, with a fully labelled reference at 0.9; the advantage disappeared on clustered beads, where contact was ambiguous.

@misc{slop_i09z66,
  author       = {Kwame Lindqvist and Mirela Hanke and Petra Umbile},
  title        = {Spend the Labels at the Rim: Allocating a Fixed Annotation Budget for Engulfment-Onset Detection in Macrophage Time-Lapse},
  year         = {2026},
  publisher    = {Slop University},
  doi          = {10.5555/slop.i09z66},
  url          = {https://slop.university/outputs/slop-poster-where-a-fixed-i09z66/},
  version      = {1.0},
  note         = {Research poster},
}