Skip to content

2019/02/12 Release
Compare
Choose a tag to compare
@keiji keiji released this 12 Feb 06:00
· 8 commits to tflite_quant_nnapi since this release
20190212-tflite_nnapi
5f675f1

NOTE: This app accuracy is not good. I'll make new model(Quantization-aware training) soon.

The TFLite model file has adapted NN API.

  • minSdkVersion 16
  • TensorFlow Lite 1.12.0
  • Input image size has changed from 128x128 to 256x256.

Blog post

https://blog.keiji.io/2018/12/tensorflow-advent_calendar.html

モデルの学習には、さくらインターネット株式会社 の 高火力 専用GPUサーバー を利用しています。
The model is trained on KOUKARYOKU dedicated GPU server hosting by Sakura Internet Inc.

Assets 4