Self-Supervised Parking Occupancy Detection Needs No Target-Lot Labels
June 17, 2026
A two-stage self-supervised approach using SimCLR with ResNet-50 achieves parking spot occupancy recognition without any labeled data from the target lot, evaluated under leave-one-out cross-environment protocol on PKLot, CNRPark-EXT, and PLds datasets.
HOW THIS AFFECTS YOU
●
researcherThe leave-one-out cross-environment evaluation protocol is a useful benchmark design for domain-generalization claims in vision tasks.