Abstract:

Automatically detecting targets of interest is very difficult when you don’t have much data. We hypothesized that Momentum Contrastive Learning could assist in lowshot object detection. We implemented code and tested this hypothesis on the xView dataset.