I’m trying to expand this existing code that defines tf.functions to be used in my tflite model in android studio. I’ve added a get weights and biases function to be able to debug an issue in my application where it seems after the first training cycle the weights stop updating.
This is the code where I’ve only added this function.
# Copyright 2021 The TensorFlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""CLI wrapper for tflite_transfer_converter.
Converts a TF model to a TFLite transfer learning model.
"""
import os
import numpy as np
import tensorflow as tf
IMG_SIZE = 224
NUM_FEATURES = 7 * 7 * 1280
NUM_CLASSES = 4
class TransferLearningModel(tf.Module):
"""TF Transfer Learning model class."""
def __init__(self, learning_rate=0.001):
"""Initializes a transfer learning model instance.
Args:
learning_rate: A learning rate for the optimzer.
"""
self.num_features = NUM_FEATURES
self.num_classes = NUM_CLASSES
# trainable weights and bias for softmax
self.ws = tf.Variable(
tf.zeros((self.num_features, self.num_classes)),
name='ws',
trainable=True)
self.bs = tf.Variable(
tf.zeros((1, self.num_classes)), name='bs', trainable=True)
# base model
self.base = tf.keras.applications.MobileNetV2(
input_shape=(IMG_SIZE, IMG_SIZE, 3),
alpha=1.0,
include_top=False,
weights='imagenet')
# loss function and optimizer
self.loss_fn = tf.keras.losses.CategoricalCrossentropy()
self.optimizer = tf.keras.optimizers.Adam(learning_rate=learning_rate)
@tf.function(input_signature=[
tf.TensorSpec([None, IMG_SIZE, IMG_SIZE, 3], tf.float32),
])
def load(self, feature):
"""Generates and loads bottleneck features from the given image batch.
Args:
feature: A tensor of image feature batch to generate the bottleneck from.
Returns:
Map of the bottleneck.
"""
x = tf.keras.applications.mobilenet_v2.preprocess_input(
tf.multiply(feature, 255))
bottleneck = tf.reshape(
self.base(x, training=False), (-1, self.num_features))
return {'bottleneck': bottleneck}
@tf.function(input_signature=[
tf.TensorSpec([None, NUM_FEATURES], tf.float32),
tf.TensorSpec([None, NUM_CLASSES], tf.float32),
])
def train(self, bottleneck, label):
"""Runs one training step with the given bottleneck features and labels.
Args:
bottleneck: A tensor of bottleneck features generated from the base model.
label: A tensor of class labels for the given batch.
Returns:
Map of the training loss.
"""
with tf.GradientTape() as tape:
logits = tf.matmul(bottleneck, self.ws) + self.bs
prediction = tf.nn.softmax(logits)
loss = self.loss_fn(prediction, label)
gradients = tape.gradient(loss, [self.ws, self.bs])
self.optimizer.apply_gradients(zip(gradients, [self.ws, self.bs]))
result = {'loss': loss}
for grad in gradients:
result[grad.name] = grad
return result
@tf.function(input_signature=[
tf.TensorSpec([None, IMG_SIZE, IMG_SIZE, 3], tf.float32)
])
def infer(self, feature):
"""Invokes an inference on the given feature.
Args:
feature: A tensor of image feature batch to invoke an inference on.
Returns:
Map of the softmax output.
"""
x = tf.keras.applications.mobilenet_v2.preprocess_input(
tf.multiply(feature, 255))
bottleneck = tf.reshape(
self.base(x, training=False), (-1, self.num_features))
logits = tf.matmul(bottleneck, self.ws) + self.bs
return {'output': tf.nn.softmax(logits)}
@tf.function(input_signature=[tf.TensorSpec(shape=[], dtype=tf.string)])
def save(self, checkpoint_path):
"""Saves the trainable weights to the given checkpoint file.
Args:
checkpoint_path: A file path to save the model.
Returns:
Map of the checkpoint file path.
"""
tensor_names = [self.ws.name, self.bs.name]
tensors_to_save = [self.ws.read_value(), self.bs.read_value()]
tf.raw_ops.Save(
filename=checkpoint_path,
tensor_names=tensor_names,
data=tensors_to_save,
name='save')
return {'checkpoint_path': checkpoint_path}
@tf.function(input_signature=[tf.TensorSpec(shape=[], dtype=tf.string)])
def restore(self, checkpoint_path):
"""Restores the serialized trainable weights from the given checkpoint file.
Args:
checkpoint_path: A path to a saved checkpoint file.
Returns:
Map of restored weight and bias.
"""
restored_tensors = {}
restored = tf.raw_ops.Restore(
file_pattern=checkpoint_path,
tensor_name=self.ws.name,
dt=np.float32,
name='restore')
self.ws.assign(restored)
restored_tensors['ws'] = restored
restored = tf.raw_ops.Restore(
file_pattern=checkpoint_path,
tensor_name=self.bs.name,
dt=np.float32,
name='restore')
self.bs.assign(restored)
restored_tensors['bs'] = restored
return restored_tensors
# Added this extra function
@tf.function(input_signature=[])
def get_weights_and_biases(self):
"""Returns the weights and biases of the head model.
Returns:
Map of weight and bias.
"""
return {'ws': self.ws, 'bs': self.bs}
@tf.function(input_signature=[])
def initialize_weights(self):
"""Initializes the weights and bias of the head model.
Returns:
Map of initialized weight and bias.
"""
self.ws.assign(tf.random.uniform((self.num_features, self.num_classes)))
self.bs.assign(tf.random.uniform((1, self.num_classes)))
return {'ws': self.ws, 'bs': self.bs}
def convert_and_save(saved_model_dir='saved_model'):
"""Converts and saves the TFLite Transfer Learning model.
Args:
saved_model_dir: A directory path to save a converted model.
"""
model = TransferLearningModel()
tf.saved_model.save(
model,
saved_model_dir,
signatures={
'load': model.load.get_concrete_function(),
'train': model.train.get_concrete_function(),
'infer': model.infer.get_concrete_function(),
'save': model.save.get_concrete_function(),
'restore': model.restore.get_concrete_function(),
'get_weights_and_biases': model.get_weights_and_biases.get_concrete_function(),
'initialize': model.initialize_weights.get_concrete_function(),
})
# Convert the model
converter = tf.lite.TFLiteConverter.from_saved_model(saved_model_dir)
converter.target_spec.supported_ops = [
tf.lite.OpsSet.TFLITE_BUILTINS, # enable TensorFlow Lite ops.
tf.lite.OpsSet.SELECT_TF_OPS # enable TensorFlow ops.
]
converter.experimental_enable_resource_variables = True
tflite_model = converter.convert()
model_file_path = os.path.join('modelnew.tflite')
with open(model_file_path, 'wb') as model_file:
model_file.write(tflite_model)
if __name__ == '__main__':
convert_and_save()
I’m printing the input and output details of the working model before i defined this function and after as you can see below with the addition of this function the input is an empty tensor which I don’t understand why.
Input Details: [{‘name’: ‘infer_feature:0’, ‘index’: 0, ‘shape’: array([ 1, 224, 224, 3]), ‘shape_signature’: array([ -1, 224, 224, 3]), ‘dtype’: <class ‘numpy.float32’>, ‘quantization’: (0.0, 0), ‘quantization_parameters’: {‘scales’: array([], dtype=float32), ‘zero_points’: array([], dtype=int32), ‘quantized_dimension’: 0}, ‘sparsity_parameters’: {}}]
arameters’: {‘scales’: array([], dtype=float32), ‘zero_points’: array([], dtype=int32), ‘quantized_dimension’: 0}, ‘sparsity_parameters’: {}}]
Input Details: []
Why does this occur and how can I fix it?