Transfer Learning Explained: Standing on the Shoulders of Pretrained Models

TLDR: You don't need millions of labeled images or months of GPU time to build a great model. Transfer learning lets you borrow a pretrained network's hard-won feature detectors, plug in a new output