Images can be ambiguous, so let's cast traditional point-representations to distributions, w/o requiring additional architecture. 👇

A Non-isotropic Probabilistic Take on Deep Metric Learning https://t.co/p3yWku7M6V

with on

20 61