Import Python models - https://www.tensorflow.org/js/tutorials/conversion/pretrained_model
tfjs-react-native - https://github.com/tensorflow/tfjs/tree/master/tfjs-react-native
Load and save model issue - https://github.com/tensorflow/tfjs/issues/8243
Transfer learning and fine-tuning - https://www.tensorflow.org/tutorials/images/transfer_learning
Tensorflow Save and Load model - https://www.tensorflow.org/tutorials/keras/save_and_load
Tensorflow Converter - https://github.com/tensorflow/tfjs/tree/master/tfjs-converter
A layers model converted from demo_model_TL_fine_tuning.h5 - https://cysun.org/public/layers_model/model.json
A more up-to-date layers model converted from base_mobilenetv2.keras - https://cysun.org/public/layers_model2/model.json
***
The current problem is likely caused by the Async Storage limit: 6MB total (default) and 2MB per entry (can't be increased). The model size is about 14MB with each shard (i.e. the bin file) about 4MB.
***
There are three Python model format: .keras (the new format), .tf (the SavedModel format), and .h5 (the old Keras format)
A Python model can be converted to a Layered Model or a Graph Model for TensorFlow.js to use.
Currently we have: .h5 -> Graph Model and it doesn't work
Can we try:
(*) .h5 -> Layered Model: just take the existing .h5, try converting it to Layered using the --output-format=tfjs_layers_model.tf -> Layered Model.keras -> Layered Model