How to perform transfer learning on .tflite models

Question

Currently, I'm trying to make an Android food recognition application. I manage to build an application to process the image and run on the "mobilenet_quant_v1_224.tflite" pretrained model. Everything works fine and now I want to add call of my own to the model to predict.

I do know that TFLITE is still in its preliminary stage, maybe some manage to break the barrier. If so, please help. Any reply/advise/approach would be helpful because I've no idea how to proceed.

Even if there's aren't any way to go around with it, please suggest me a way to achieve the aim of this application.

Link to the trained model : http://download.tensorflow.org/models/mobilenet_v1_2018_02_22/mobilenet_v1_1.0_224_quant.tgz

Really hope to get some help.

Again Thanks in Advance.

Answer 1

As of now, TFLite does not support training. You can do the transfer learning on the TF model, and then convert the transfer-learnt model to TFLite. This TF-for-poets-2-tflite codelab walks you through that exactly (including links to doing transfer learning on your TF model).