Tracking paper wasps in a face-learning assay
For Christopher Jernigan's Animal Cognition study of face recognition in paper wasps, I tracked wasps in a face-learning arena with DeepLabCut and measured their movement and time in each zone with SimBA's region-of-interest tool.
Christopher Jernigan’s study asked whether northern paper wasps (Polistes fuscatus) need color to tell faces apart. Each wasp learned to discriminate two wasp faces, printed either in color or in grayscale, in a small arena with one face at each end and a floor that could deliver a mild shock in front of one of them. Every training and testing trial was filmed from above. My part was the computer-vision analysis of those videos.
What I did
- Tracking with DeepLabCut. A DeepLabCut model followed each wasp’s head, thorax and abdomen tip through every trial. The paper’s measures use the thorax, the same point the experimenters used to decide when a wasp had crossed into a zone. The model was trained on 589 hand-labelled frames from 89 videos, and positions below a 0.9 likelihood were discarded.
- SimBA’s region-of-interest tool. From the thorax track, SimBA measured how far each wasp moved, its average speed and the share of each trial it spent in the zone in front of each face (the shock-paired face and the safe face).
The clip at the top is part of one unreinforced testing trial of a wasp trained on color faces: DeepLabCut’s head (blue), thorax (pink) and abdomen-tip (yellow) points.
The paper
Jernigan, C.M., Stafstrom, J.A., Zaba, N.C., Vogt, C.C., & Sheehan, M.J. (2022). Color is necessary for face discrimination in the Northern paper wasp, Polistes fuscatus. Animal Cognition. doi:10.1007/s10071-022-01691-9
Read the paper for what the wasps learned and what it says about how they recognize faces.
Thank you to Christopher Jernigan, who led this study.