using .json pretraind model in matlab

i download pretrained model from this url: https://tfhub.dev/mediapipe/tfjs-model/handpose_3d/detector/full/1?tfjs-format=compressed

and now i have two files: 1)model.json 2)group1-shard1of1.bin

how can i use this pretrined model in matlab

or convert this pretrained model to .h5 model

Answer 1

[Shreeya]

Hi Amir

You can import a pretrained JSON/h5 model in matlab as a neural network using the "importKerasNetwork".The argument passed to the function is the file name of the model you wish to import. It returns a pretrained keras neural network which can be used for further prediction tasks.

Refer to the documentaion for more details: https://in.mathworks.com/help/deeplearning/ref/importkerasnetwork.html

Hope this helps.

[Amir Azadeh Ranjbar]

thanks.but not work.... download model from : https://tfhub.dev/mediapipe/tfjs-model/handskeleton/1/default/1

[Shreeya]

Try passing in the .bin file as the 'WeightFile' argument.

[Amir Azadeh Ranjbar]

modelfile = 'model.json';

weights = 'group1-shard1of2.bin';

net = importKerasNetwork(modelfile,'WeightFile',weights, ...

'OutputLayerType','classification')

Error using nnet.internal.cnn.keras.importKerasNetwork

Error reading Keras model_config from file 'model.json'. The error message

was: 'Unrecognized field name "keras_version".''