Unable to use models created by the tensorflowjs converter

Question

I am trying to reuse with tensorflowjs models created by tensorflow. In order to understand how the converter works I have tried to convert the mobilenetv2 model:

tensorflowjs_converter --input_format=tf_hub --output_format=tensorflowjs   'https://tfhub.dev/google/imagenet/mobilenet_v2_050_224/classification/2' ./web_model

That seems to work. Then I have tried to used this new converted model within the mobilenet demo by changing the way the model is loaded:

// const model = await mobilenet.load({version, alpha});
// replaced by
const model = await mobilenet.load({ modelUrl: './web_model/model.json', version, alpha, inputRange: [0, 1], fromTFHub: true });

// Classify the image.
const predictions = await model.classify(img);

The classify call triggers an error:

Uncaught (in promise) Error: Activation relu6 has not been implemented for the WebGL backend.

I have no clue on how the official tensorflowjs mobilenet model has been generated :(

edkeveked edkeveked · Accepted Answer · 2019-09-13T12:28:29

from keras.applications import MobileNetV2
model = MobileNetV2(weights='imagenet', include_top=False)

save_model(
    model,
    "mobilenet2.h5",
    overwrite=True,
)

Convert mobilenet feature extractor to js

tensorflowjs_converter --input_format keras \
                       path/to/mobilenet2.h5 \
                       path/to/tfjs_target_dir

Unable to use models created by the tensorflowjs converter

3 Answers