The official "Edge TPU model" has been released.
If you want to create your own model, see this link (Quantize DeepLab model for faster on-device inference).
Be careful with the compiler and library versions. Previous versions do not work properly.
- Edge TPU Compiler version 2.0.267685300
- TPU Python library virsion 2.12.1 (September 2019)