deep image matting

import keras2onnx
import onnx

if __name__ == '__main__':
    img_rows, img_cols = 320, 320
    channel = 4

    pretrained_path = 'models/final.42-0.0398.hdf5'
    encoder_decoder = build_encoder_decoder()
    final = build_refinement(encoder_decoder)
    final.load_weights(pretrained_path)
    print(final.summary())

    onnx_model = keras2onnx.convert_keras(final, final.name, target_opset=10)
    temp_model_file = 'deep-image-matting.onnx'
    onnx.save_model(onnx_model, temp_model_file)