1 paper
Artheme Gauthier-Villar, Guodong Ding, Angela Yao
Dataset condensation for action segmentation synthesizes compact, informative representations of long, untrimmed video datasets. The existing approach relies on Variational Autoenc…