which is the version of tensorflow in this project？

[xieshenru]

pip insatll tensorflow==2.0.0 or pip install tensorflow==2.0.0-beta1，tensorflow2.0.0or tensorflow2.0.0-beta1,but when I run train.py, the error of "ValueError: This converter can only convert a single ConcreteFunction. Converting multiple functions is under development" appears,why?How to do?

[yeyupiaoling]

Tensorflow >= 2.0.0

[xieshenru]

When runing the line of converting to tflite in train.py,"the error of "ValueError: This converter can only convert a single ConcreteFunction. Converting multiple functions is under development"appears,I can't convert to tflite,the version of tensorflow is TF2.0,why?You can convert to tflite?Any other ways to convert to tflite?Thanks!

[yeyupiaoling]

@xieshenru first, train your model:

# save model
tf.keras.models.save_model(model=model, filepath=cfg.H5_MODEL_PATH, save_format='h5')

then convert tflite model

import tensorflow as tf
import config as cfg
import reader

# load h5 model
model = tf.keras.models.load_model(cfg.H5_MODEL_PATH)

# convert tflite model
converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()
open(cfg.TFLITE_MODEL_FILE, 'wb').write(tflite_model)
print('saved tflite model!')

[xieshenru]

wen I use "tf.keras.models.save_model(model=model, filepath=cfg.H5_MODEL_PATH, save_format='h5') to save model, the error of "NotImplementedError: Saving the model to HDF5 format requires the model to be a Functional model or a Sequential model. It does not work for subclassed models, because such models are defined via the body of a Python method, which isn't safely serializable. Consider saving to the Tensorflow SavedModel format (by setting save_format="tf") or using save_weights."How to solve this problem?Thanks for your help！

[yeyupiaoling]

@xieshenru Are you traning SSD mode? I used those codes to converter mobilenetV2 model.

[yeyupiaoling]

@xieshenru You can try this code, when you train model.

            converter = tf.lite.TFLiteConverter.from_keras_model(model)
            tflite_model = converter.convert()
            open('model/model.tflite', 'wb').write(tflite_model)
            print('saved tflite model!')

[xieshenru]

I used those codes to convert ShuffleNetV2 and mobilenetV2 model for Classification problem，use the code of train.py ,I can't convert to tflit.Here is the code used in your "train.py"

tf.saved_model.save(model, save_model_dir)

convert to tensorflow lite format

converter = tf.lite.TFLiteConverter.from_saved_model(save_model_dir)
 tflite_model = converter.convert()
open("converted_model.tflite", "wb").write(tflite_model)

Can you convert shuffleNetv2 or mobilenetv2 to tflite by using the code of "train.py"?

[yeyupiaoling]

@xieshenru I can save the tflite model in the train code I wrote.

[calmiLovesAI]

You can try to use the following code to convert to tflite.

converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()
open(TFLite_model_dir, "wb").write(tflite_model)

I will update the code in the next few days.

[yeyupiaoling]

@calmisential Can you answer this issue? https://github.com/calmisential/TensorFlow2.0_SSD/issues/8

[calmiLovesAI]

@calmisential Can you answer this issue? calmisential/TensorFlow2.0_SSD#8

It will take some time to find a solution, and I'm working on it.

[calmiLovesAI]

pip insatll tensorflow==2.0.0 or pip install tensorflow==2.0.0-beta1，tensorflow2.0.0or tensorflow2.0.0-beta1,but when I run train.py, the error of "ValueError: This converter can only convert a single ConcreteFunction. Converting multiple functions is under development" appears,why?How to do?

I have updated the version of tensorflow to 2.1.0, the issue has been resolved in the latest project code.

[xieshenru]

@calmisential @yeyupiaoling Thanks for your help! I can try to use the following code from you to convert to lite .

converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()

open(TFLite_model_dir, "wb").write(tflite_model)

but when I use the mode of model.tflite by python to predict the image which be input,the error of "RuntimeError: tensorflow/lite/kernels/transpose.cc Transpose op only supports 1D-4D input arrays.Node number 9 (TRANSPOSE) failed to prepare." in the line of "interpreter.allocate_tensors()". The following code is which I use.

    interpreter = tf.lite.Interpreter(model_path=model_path)
    interpreter.allocate_tensors()
    # Get input and output tensors.
    input_details = interpreter.get_input_details()
    output_details = interpreter.get_output_details()
    full_path = os.path.join(PATH_TEST_IMAGES, filename)
    img = cv2.imread(full_path)
    img = cv2.resize(img, (160, 160))
    image_np_expanded = np.expand_dims(img, axis=0)
    image_np_expanded = image_np_expanded.astype('float32') 
    interpreter.set_tensor(input_details[0]['index'], image_np_expanded)
    interpreter.invoke()
    output_data = interpreter.get_tensor(output_details[0]['index'])

I don't know why this problem occurs, do you know how to solve it?Thanks!