Unhandled "Dequantize" operation in new MediaPipe BlazeFace model about pinto_model_zoo

[PINTO0309]

I have created a conversion tool that supports Dequantize operation (built-in_code = 6). I just created it 3 days ago, so I will add more OPs to support it.
https://github.com/PINTO0309/tflite2tensorflow

[PINTO0309]

The weights set in each layer were extracted by hand using Netron. It is possible to extract the weights in bulk if you create a script, but so far I have not created such a script.

The script to convert the new model is below. All you have to do is specify the weight file in Numpy format extracted from the above work in each layer in np.load. I put all the weight files under a folder named weights_front and converted them.
https://github.com/PINTO0309/PINTO_model_zoo/blob/master/030_BlazeFace/01_float32/07_new_blazeface_front_tflite2h5_weight_int_fullint_float16_quant.py

PINTO_model_zoo/030_BlazeFace/01_float32/07_new_blazeface_front_tflite2h5_weight_int_fullint_float16_quant.py

Lines 44 to 46 in c50411c

	conv1_1 = Conv2D(filters=24, kernel_size=[5, 5], strides=[2, 2], padding="same", dilation_rate=[1, 1], activation='relu',
	kernel_initializer=Constant(np.load('weights_front/conv2d_Kernel').transpose(1,2,3,0)),
	bias_initializer=Constant(np.load('weights_front/conv2d_Bias')))(inputs)

[PINTO0309]

This is closed as it has been resolved. If you have other issues, please post a new issue.