我已经训练了一个非常简单的模型并将其转换为Tflite模型....该模型的python代码如下
# -*- coding: utf-8 -*-
"""
Created on Sat Sep 28 21:05:22 2019
@author: Aneshka Goyal
"""
import tensorflow as tf
def freeze_session(session, keep_var_names=None, output_names=None, clear_devices=True):
"""
Freezes the state of a session into a pruned computation graph.
Creates a new computation graph where variable nodes are replaced by
constants taking their current value in the session. The new graph will be
pruned so subgraphs that are not necessary to compute the requested
outputs are removed.
@param session The TensorFlow session to be frozen.
@param keep_var_names A list of variable names that should not be frozen,
or None to freeze all the variables in the graph.
@param output_names Names of the relevant graph outputs.
@param clear_devices Remove the device directives from the graph for better portability.
@return The frozen graph definition.
"""
graph = session.graph
with graph.as_default():
freeze_var_names = list(set(v.op.name for v in tf.global_variables()).difference(keep_var_names or []))
output_names = output_names or []
output_names += [v.op.name for v in tf.global_variables()]
input_graph_def = graph.as_graph_def()
if clear_devices:
for node in input_graph_def.node:
node.device = ""
frozen_graph = tf.graph_util.convert_variables_to_constants(
session, input_graph_def, output_names, freeze_var_names)
return frozen_graph
inp = tf.placeholder(name="inp", dtype=tf.float32, shape=(1, 1))
w = tf.Variable(tf.zeros([1, 1], tf.float32), dtype=tf.float32, name="w")
y = tf.matmul(w, inp)
out = tf.identity(y, name="out")
init_op = tf.global_variables_initializer()
with tf.Session() as sess:
sess.run(init_op)
# After Init var, change the value to 2
assignment = w.assign([[2]])
sess.run(assignment)
output = sess.run(out, feed_dict={inp: [[1]]})
print (output)
frozen_graph = freeze_session(sess, output_names=[out.op.name])
tflite_model = tf.contrib.lite.toco_convert(frozen_graph, [inp], [out])
open("mat_mul.tflite", "wb").write(tflite_model)
我能够获得此模型的输入和输出的正确尺寸，并将其放入以下android代码中，以便与我的android应用集成，但是我怀疑从FirebaseInterpreter..as中正确取出输出通过对try-catch块的注释指示。
package com.example.aneshkagoyal.samplecustom;
import android.support.annotation.NonNull;
import android.support.v7.app.AppCompatActivity;
import android.os.Bundle;
import android.util.Log;
import android.widget.TextView;
import android.widget.Toast;
import com.google.android.gms.tasks.OnFailureListener;
import com.google.android.gms.tasks.OnSuccessListener;
import com.google.firebase.FirebaseApp;
import com.google.firebase.ml.common.FirebaseMLException;
import com.google.firebase.ml.custom.FirebaseModelDataType;
import com.google.firebase.ml.custom.FirebaseModelInputOutputOptions;
import com.google.firebase.ml.custom.FirebaseModelInputs;
import com.google.firebase.ml.custom.FirebaseModelInterpreter;
import com.google.firebase.ml.custom.FirebaseModelManager;
import com.google.firebase.ml.custom.FirebaseModelOptions;
import com.google.firebase.ml.custom.FirebaseModelOutputs;
import com.google.firebase.ml.custom.model.FirebaseLocalModelSource;
public class MainActivity extends AppCompatActivity {
FirebaseModelInterpreter firebaseInterpreter;
FirebaseModelInputs inputs;
FirebaseModelInputOutputOptions inputOutputOptions;
TextView t;
@Override
protected void onCreate(Bundle savedInstanceState) {
super.onCreate(savedInstanceState);
setContentView(R.layout.activity_main);
t = findViewById(R.id.my_text);
FirebaseLocalModelSource localSource =
new FirebaseLocalModelSource.Builder("mat_mul") // Assign a name to this model
.setAssetFilePath("mat_mul.tflite")
.build();
FirebaseModelManager.getInstance().registerLocalModelSource(localSource);
FirebaseModelOptions options = new FirebaseModelOptions.Builder()
.setLocalModelName("asset")
.build();
try {
firebaseInterpreter =
FirebaseModelInterpreter.getInstance(options);
} catch (FirebaseMLException e) {
e.printStackTrace();
}
// Define the Input and Output dimensions and types
try {
inputOutputOptions = new FirebaseModelInputOutputOptions.Builder()
.setInputFormat(0, FirebaseModelDataType.FLOAT32, new int[]{1,1})//This line
.setOutputFormat(0, FirebaseModelDataType.FLOAT32, new int[]{1,1})//This line
.build();
} catch (FirebaseMLException e) {
e.printStackTrace();
}
float[][] input;
input = new float[1][1];
//output = new float[1][1];
input[0][0]= 21.0f;
try {
inputs = new FirebaseModelInputs.Builder()
.add(input) // add() as many input arrays as your model requires
.build();
} catch (FirebaseMLException e) {
e.printStackTrace();
Log.d("Failure","in InputBuilder");
}
//Check if the output is being correctly taken as float[1][1]
try {
firebaseInterpreter.run(inputs, inputOutputOptions)
.addOnSuccessListener(
new OnSuccessListener<FirebaseModelOutputs>() {
@Override
public void onSuccess(FirebaseModelOutputs result) {
Toast.makeText(MainActivity.this,"output"+result.<float[][]>getOutput(0).toString(),Toast.LENGTH_SHORT).show();
Log.d("ANSWERIS", result.<float[][]>getOutput(0).toString());
t.setText(result.<float[][]>getOutput(0).toString());
}
})
.addOnFailureListener(
new OnFailureListener() {
@Override
public void onFailure(@NonNull Exception e) {
// Task failed with an exception
// ...
Log.d("Failure","ho gyaa");
}
});
} catch (FirebaseMLException e) {
e.printStackTrace();
}
}
}
我在应用程序中获得的输出是值[[F @ 6061395]的垃圾分类，并且该值每次都会更改，并且运行python脚本（仅）时获得的输出为[[2.]]，应为是正确的答案。 请根据我的感觉建议可能的错误源，以帮助解决问题，因为我不确定我是否做对了。