This implenment is based on Google Dev Board There are two kind of Transfer Learning on Dev Board - Backpropogation and Imprinted weight. You can use the slide below to implentment easily.
Update Dev Board Python library
Make sure that the Python library has already updated to 2.11.1. Updated Edge TPU Python library: https://coral.withgoogle.com/news/updates-07-2019/.
Retrain the classification model by backpropogation. Backpropagation will update the in weights in every where. If you use this way, the accuracy will be higher than Imprinted weight.
Weight imprinting is a technique for retraining a classification models using a small set of sample data. It's based on : Low-Shot Learning with Imprinted Weights https://arxiv.org/pdf/1712.07136.pdf. Weight Impringting require very few sample images (fewer than 10 training samples can achieve high accuracy). Nevertheless, it has difficulty learning from datasets with large intra-class variation.If your use-case expects data with high intra-class variance, consider instead using on-device transfer learning with backpropagation
Slide with implement detail
Implement Transfer Learning on Dev Board: https://drive.google.com/open?id=16TA87fefz00IRBdtywsvjkbUp5M6-wWJ
Google official website: https://coral.withgoogle.com/docs/edgetpu/retrain-classification-ondevice/
Low-Shot Learning with Imprinted Weights: https://arxiv.org/pdf/1712.07136.pdf
Updated Edge TPU Python library: https://coral.withgoogle.com/news/updates-07-2019/