Keras pretrain weight

may I get this weight please?
thanks :kissing_smiling_eyes:

Hi @LenBonJJ, the models are now here: Just append the /keras/mobilenet etc part there.

We’re fixing this next week in the notebooks.

may I know your quantization way?
I am trying to quantized and used in openmv but failed
and this is my quantization code and error:

I will appreciate it if you can give me some advice :smile:

by the way, I used this pretrain model:“mobilenet_v2_weights_tf_dim_ordering_tf_kernels_0.35_96.h5”

@LenBonJJ this is the code we use:

        print('Converting TensorFlow Lite int8 quantized model with int8 input and output...', flush=True)
        converter_quantize = tf.lite.TFLiteConverter.from_concrete_functions([concrete_func])
        converter_quantize.optimizations = [tf.lite.Optimize.DEFAULT]
        converter_quantize.representative_dataset = dataset_generator
        # Force the input and output to be int8
        converter_quantize.target_spec.supported_ops = [tf.lite.OpsSet.TFLITE_BUILTINS_INT8]
        converter_quantize.inference_input_type = tf.int8
        converter_quantize.inference_output_type = tf.int8
        tflite_quant_model = converter_quantize.convert()
        open(os.path.join(dir_path, filename), 'wb').write(tflite_quant_model)
        return tflite_quant_model

Not sure what’s going on with your model, but hard to give support outside of the Studio (where it just works :wink: )

oh thanks
I think I know where the problem is :thinking:
may I know how can you get the concrete_func?
Thank you very much! :smile:

Hi @LenBonJJ it’s actually part of TensorFlow. See