问题

I will use yolo weights in android so I plan to convert yolo weights file to tflite file.

I use this code in anaconda prompt because I downloaded keras library in env.

activate env   
python convert.py yolov3.cfg yolov3.weights model_data/yolo.h5

Finally, it did.Saved Keras model to model_data/yolo.h5

And I'm going to convert this h5 file to tflite file in jupyter notebook with this code.

model = tf.keras.models.load_model("./yolo/yolo.h5", compile=False)
converter = tf.lite.TFLiteConverter.from_keras_model(model)
tflite_model = converter.convert()
open("keras_model.tflite", "wb").write(tflite_model)

But this error occurs.

ValueError                                Traceback (most recent call last)

<ipython-input-3-964a59978091> in <module>()

  1 model = tf.keras.models.load_model("./yolo/yolo.h5", compile=False)

  2 converter = tf.lite.TFLiteConverter.from_keras_model(model)

----> 3 tflite_model = converter.convert()

  4 open("keras_model.tflite", "wb").write(tflite_model)



~\Anaconda3\envs\tensorflow\lib\site-packages\tensorflow_core\lite\python\lite.py in convert(self)

426         raise ValueError(

427             "None is only supported in the 1st dimension. Tensor '{0}' has "

--> 428             "invalid shape '{1}'.".format(_get_tensor_name(tensor), shape_list))

429       elif shape_list and shape_list[0] is None:

430         # Set the batch size to 1 if undefined.



ValueError: None is only supported in the 1st dimension. Tensor 'input_1' has invalid shape '[None, None, None, 3]'.

How can I fix it?

our model summary is

Model: "model_1"

Layer (type) Output Shape Param # Connected to

input_1 (InputLayer) [(None, None, None, 0

conv2d_1 (Conv2D) (None, None, None, 3 864 input_1[0][0]

batch_normalization_1 (BatchNor (None, None, None, 3 128 conv2d_1[0][0]

leaky_re_lu_1 (LeakyReLU) (None, None, None, 3 0 batch_normalization_1[0][0]

zero_padding2d_1 (ZeroPadding2D (None, None, None, 3 0 leaky_re_lu_1[0][0]

conv2d_2 (Conv2D) (None, None, None, 6 18432 zero_padding2d_1[0][0]

batch_normalization_2 (BatchNor (None, None, None, 6 256 conv2d_2[0][0]

leaky_re_lu_2 (LeakyReLU) (None, None, None, 6 0 batch_normalization_2[0][0]

conv2d_3 (Conv2D) (None, None, None, 3 2048 leaky_re_lu_2[0][0]

. . . .

batch_normalization_65 (BatchNo (None, None, None, 5 2048 conv2d_66[0][0]

batch_normalization_72 (BatchNo (None, None, None, 2 1024 conv2d_74[0][0]

leaky_re_lu_58 (LeakyReLU) (None, None, None, 1 0 batch_normalization_58[0][0]

leaky_re_lu_65 (LeakyReLU) (None, None, None, 5 0 batch_normalization_65[0][0]

leaky_re_lu_72 (LeakyReLU) (None, None, None, 2 0 batch_normalization_72[0][0]

conv2d_59 (Conv2D) (None, None, None, 2 261375 leaky_re_lu_58[0][0]

conv2d_67 (Conv2D) (None, None, None, 2 130815 leaky_re_lu_65[0][0]

conv2d_75 (Conv2D) (None, None, None, 2 65535 leaky_re_lu_72[0][0]

Total params: 62,001,757 Trainable params: 61,949,149 Non-trainable params: 52,608

回答1:

I see that you're getting the H5 file of the Keras YOLO model. For TFLite models, you'll require a model that has a definite input shape like ( 256 , 256 , 3 ). Also, for an H5 model, you can't modify the input shape after the model is saved. So, you can take these measures,

Go to the file where the model and its layers are instantiated, modify the input shape there. Retrain the model using that fixed input size.
Use the TFLite Object Detection API. This has an Android sample app too.
Train the model using DarkNet as mentioned here.
If you're trying to detect one of the 20 classes present in the PASCAL, use this model.