# tf-lite-demo

**Repository Path**: deng-yongsheng/tf-lite-demo

## Basic Information

- **Project Name**: tf-lite-demo
- **Description**: TensorflowLite Demo：拟合sin函数，并将模型移植到C语言下
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2022-10-26
- **Last Updated**: 2022-10-26

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# TensorflowLite Demo：拟合sin函数，并将模型移植到C语言下

## keras训练模型拟合sin函数


```python
import keras
from keras.layers import *
from keras.models import Sequential
import numpy as np
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
```

### 使用numpy生成sin函数的x和y映射


```python
x = np.linspace(0, 1, 10000, dtype='float32')
y = (np.sin(x * 2 * np.pi) + 1) * 0.5
```

### 绘制`x`, `y`图像


```python
plt.figure(dpi=120)
plt.plot(x, y)
```

![png](md_files/1.keras%E6%A8%A1%E5%9E%8B%E6%9E%84%E5%BB%BA%E5%92%8C%E8%AE%AD%E7%BB%83_5_1.png)


### 将x,y矩阵转换，变成训练数据，


```python
x_train = x.reshape((-1,1))
y_train = y.reshape((-1,1))
```

### 创建模型


```python
model = Sequential([
    InputLayer((1,)),
    Dense(32,activation='relu'),
    Dense(64,activation='relu'),
    Dense(1,activation='sigmoid'),
])
model.compile(optimizer='adam',
             loss='mse',
             metrics=['accuracy'])
model.summary()
```

    Model: "sequential"
    _________________________________________________________________
    Layer (type)                 Output Shape              Param #   
    =================================================================
    dense (Dense)                (None, 32)                64        
    _________________________________________________________________
    dense_1 (Dense)              (None, 64)                2112      
    _________________________________________________________________
    dense_2 (Dense)              (None, 1)                 65        
    =================================================================
    Total params: 2,241
    Trainable params: 2,241
    Non-trainable params: 0
    _________________________________________________________________


### 模型训练


```python
model.fit(x_train, y_train, batch_size=10000, epochs=10000)
```

### 模型测试


```python
predict_y = model.predict(x)
```

### 拟合结果绘制图像


```python
plt.figure(dpi=72*2)
plt.title('Neural network fitting sin function')
plt.plot(x, y, color='b', label='sin')
plt.plot(x, predict_y, color='r', label='fitted')
plt.legend(loc=0)
```

![png](md_files/1.keras%E6%A8%A1%E5%9E%8B%E6%9E%84%E5%BB%BA%E5%92%8C%E8%AE%AD%E7%BB%83_15_1.png)
​    


### 模型保存到文件


```python
model.save('sin.model.h5')
```


## 将keras模型转换成tflite模型


```python
import keras
import tensorflow as tf
```

### 从文件中加载keras模型


```python
model = keras.models.load_model('sin.model.h5')
model.summary()
```

    Model: "sequential"
    _________________________________________________________________
    Layer (type)                 Output Shape              Param #   
    =================================================================
    dense (Dense)                (None, 32)                64        
    _________________________________________________________________
    dense_1 (Dense)              (None, 64)                2112      
    _________________________________________________________________
    dense_2 (Dense)              (None, 1)                 65        
    =================================================================
    Total params: 2,241
    Trainable params: 2,241
    Non-trainable params: 0
    _________________________________________________________________


### 从keras模型转换到tflite模型


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

    INFO:tensorflow:Assets written to: C:\Users\sheng\AppData\Local\Temp\tmpkxydx7qq\assets


### 将转换后的模型写入文件


```python
with open('sin.tf_lite.model','wb') as f:
    f.write(tflite_model)
```

## tflite部署模型


```python
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
```

### 从文件加载解释器


```python
interpreter = tf.lite.Interpreter('sin.tf_lite.model')
interpreter.allocate_tensors()
```

### 获取模型的输入输出信息


```python
interpreter.get_output_details()
interpreter.get_input_details()
```


    [{'name': 'input_1',
      'index': 0,
      'shape': array([1, 1]),
      'shape_signature': array([-1,  1]),
      'dtype': numpy.float32,
      'quantization': (0.0, 0),
      'quantization_parameters': {'scales': array([], dtype=float32),
       'zero_points': array([], dtype=int32),
       'quantized_dimension': 0},
      'sparsity_parameters': {}}]



### 生成测试数据


```python
x = np.linspace(0,1,num=1000,dtype='float32')
```

### 获取输入和输出tensor的索引


```python
input_tensor_index = interpreter.get_input_details()[0]['index']
output_tensor_index = interpreter.get_output_details()[0]['index']
```

### 设置输入tensor


```python
interpreter.set_tensor(input_tensor_index, x[0].reshape((-1,1)))
```

### 调用解释器


```python
interpreter.invoke()
```

### 获取输出tensor


```python
interpreter.get_tensor(output_tensor_index)
```


    array([[0.49774837]], dtype=float32)



## 批量预测


```python
y= []
for i in x:
    interpreter.set_tensor(input_tensor_index, i.reshape((1,1)))
    interpreter.invoke()
    y.append(interpreter.get_tensor(output_tensor_index)[0][0])
```


```python
plt.plot(x,y)
```


![png](md_files/3.tflite%E9%83%A8%E7%BD%B2%E6%A8%A1%E5%9E%8B_18_1.png)