# Resnet.py
#!/usr/bin/env python
# -*- coding:utf-8 -*-
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, Sequential
class BasicBlock(layers.Layer):

def init(self, filter_num, stride=1):

super(BasicBlock, self).init()
    self.conv1 = layers.Conv2D(filter_num, (<span class="hljs-number">3</span>, <span class="hljs-number">3</span>), strides=stride, padding=<span class="hljs-string">‘same‘</span>)
    self.bn1 = layers.BatchNormalization()
    self.relu = layers.Activation(<span class="hljs-string">‘relu‘</span>)

    self.conv2 = layers.Conv2D(filter_num, (<span class="hljs-number">3</span>, <span class="hljs-number">3</span>), strides=<span class="hljs-number">1</span>, padding=<span class="hljs-string">‘same‘</span>)
    self.bn2 = layers.BatchNormalization()

    <span class="hljs-keyword">if</span> stride != <span class="hljs-number">1</span>:
        self.downsample = Sequential()
        self.downsample.add(layers.Conv2D(filter_num, (<span class="hljs-number">1</span>, <span class="hljs-number">1</span>), strides=stride))
    <span class="hljs-keyword">else</span>:
        self.downsample = <span class="hljs-keyword">lambda</span> x: x

<span class="hljs-function"><span class="hljs-keyword">def</span> <span class="hljs-title">call</span>(<span class="hljs-params">self, inputs, training=<span class="hljs-literal">None</span></span>):</span>
    <span class="hljs-comment"># [b,h,w,c]</span>
    out = self.conv1(inputs)
    out = self.bn1(out)
    out = self.relu(out)

    out = self.conv2(out)
    out = self.bn2(out)

    identity = self.downsample(inputs)

    output = layers.add([out, identity])
    output = tf.nn.relu(output)

    <span class="hljs-keyword">return</span> out

Res Block

ResNet18

# Resnet.py
#!/usr/bin/env python
# -*- coding:utf-8 -*-
import tensorflow as tf
from tensorflow import keras
from tensorflow.keras import layers, Sequential
class BasicBlock(layers.Layer):

def init(self, filter_num, stride=1):

super(BasicBlock, self).init()
    self.conv1 = layers.Conv2D(filter_num, (<span class="hljs-number">3</span>, <span class="hljs-number">3</span>), strides=stride, padding=<span class="hljs-string">‘same‘</span>)
    self.bn1 = layers.BatchNormalization()
    self.relu = layers.Activation(<span class="hljs-string">‘relu‘</span>)

    self.conv2 = layers.Conv2D(filter_num, (<span class="hljs-number">3</span>, <span class="hljs-number">3</span>), strides=<span class="hljs-number">1</span>, padding=<span class="hljs-string">‘same‘</span>)
    self.bn2 = layers.BatchNormalization()

    <span class="hljs-keyword">if</span> stride != <span class="hljs-number">1</span>:
        self.downsample = Sequential()
        self.downsample.add(layers.Conv2D(filter_num, (<span class="hljs-number">1</span>, <span class="hljs-number">1</span>), strides=stride))
    <span class="hljs-keyword">else</span>:
        self.downsample = <span class="hljs-keyword">lambda</span> x: x

<span class="hljs-function"><span class="hljs-keyword">def</span> <span class="hljs-title">call</span>(<span class="hljs-params">self, inputs, training=<span class="hljs-literal">None</span></span>):</span>
    <span class="hljs-comment"># [b,h,w,c]</span>
    out = self.conv1(inputs)
    out = self.bn1(out)
    out = self.relu(out)

    out = self.conv2(out)
    out = self.bn2(out)

    identity = self.downsample(inputs)

    output = layers.add([out, identity])
    output = tf.nn.relu(output)

    <span class="hljs-keyword">return</span> out

class ResNet(keras.Model):

def init(self, layer_dims, num_classes=100):  # [2,2,2,2]

super(ResNet, self).init()
    <span class="hljs-comment"># 根部</span>
    self.stem = Sequential([layers.Conv2D(<span class="hljs-number">64</span>, (<span class="hljs-number">3</span>, <span class="hljs-number">3</span>), strides=(<span class="hljs-number">1</span>, <span class="hljs-number">1</span>,)),
                            layers.BatchNormalization(),
                            layers.Activation(<span class="hljs-string">‘relu‘</span>),
                            layers.MaxPool2D(pool_size=(<span class="hljs-number">2</span>, <span class="hljs-number">2</span>), strides=(<span class="hljs-number">1</span>, <span class="hljs-number">1</span>), padding=<span class="hljs-string">‘same‘</span>)
                            ])

    <span class="hljs-comment"># 64,128,256,512是通道数</span>
    self.layer1 = self.build_resblock(<span class="hljs-number">64</span>, layer_dims[<span class="hljs-number">0</span>])
    self.layer2 = self.build_resblock(<span class="hljs-number">128</span>, layer_dims[<span class="hljs-number">1</span>], stride=<span class="hljs-number">2</span>)
    self.layer3 = self.build_resblock(<span class="hljs-number">256</span>, layer_dims[<span class="hljs-number">2</span>], stride=<span class="hljs-number">2</span>)
    self.layer4 = self.build_resblock(<span class="hljs-number">512</span>, layer_dims[<span class="hljs-number">3</span>], stride=<span class="hljs-number">2</span>)

    <span class="hljs-comment"># output: [b, 512, h, w]</span>
    self.avgpool = layers.GlobalAveragePooling2D()
    self.fc = layers.Dense(num_classes)  <span class="hljs-comment"># 分类</span>

<span class="hljs-function"><span class="hljs-keyword">def</span> <span class="hljs-title">call</span>(<span class="hljs-params">self, inputs, training=<span class="hljs-literal">None</span></span>):</span>
    x = self.stem(inputs)

    x = self.layer1(x)
    x = self.layer2(x)
    x = self.layer3(x)
    x = self.layer4(x)

    <span class="hljs-comment"># [b, c]</span>
    x = self.avgpool(x)
    <span class="hljs-comment"># [b]</span>
    x = self.fc(x)

    <span class="hljs-keyword">return</span> x

<span class="hljs-function"><span class="hljs-keyword">def</span> <span class="hljs-title">build_resblock</span>(<span class="hljs-params">self, filter_num, blocks, stride=<span class="hljs-number">1</span></span>):</span>
    res_blocks = Sequential()
    <span class="hljs-comment"># may down sample</span>
    res_blocks.add(BasicBlock(filter_num, stride))

    <span class="hljs-keyword">for</span> _ <span class="hljs-keyword">in</span> <span class="hljs-built_in">range</span>(<span class="hljs-number">1</span>, blocks):
        res_blocks.add(BasicBlock(filter_num, stride=<span class="hljs-number">1</span>))

    <span class="hljs-keyword">return</span> res_blocks

def resnet18():

return ResNet([2, 2, 2, 2])
def resnet34():

return ResNet([3, 4, 6, 3])

# resnet18_train.py
#!/usr/bin/env python
# -*- coding:utf-8 -*-
import tensorflow as tf
from tensorflow.keras import layers, optimizers, datasets, Sequential
import os
from Resnet import resnet18

os.environ[‘TF_CPP_MIN_LOG_LEVEL‘] = ‘2‘
tf.random.set_seed(2345)


def preprocess(x, y):
    # [-1~1]
    x = tf.cast(x, dtype=tf.float32) / 255. - 0.5
    y = tf.cast(y, dtype=tf.int32)
    return x, y


(x, y), (x_test, y_test) = datasets.cifar100.load_data()
y = tf.squeeze(y, axis=1)
y_test = tf.squeeze(y_test, axis=1)
print(x.shape, y.shape, x_test.shape, y_test.shape)

train_db = tf.data.Dataset.from_tensor_slices((x, y))
train_db = train_db.shuffle(1000).map(preprocess).batch(512)

test_db = tf.data.Dataset.from_tensor_slices((x_test, y_test))
test_db = test_db.map(preprocess).batch(512)

sample = next(iter(train_db))
print(‘sample:‘, sample[0].shape, sample[1].shape,
      tf.reduce_min(sample[0]), tf.reduce_max(sample[0]))


def main():
    # [b, 32, 32, 3] => [b, 1, 1, 512]
    model = resnet18()
    model.build(input_shape=(None, 32, 32, 3))
    model.summary()
    optimizer = optimizers.Adam(lr=1e-3)

    for epoch in range(500):

        for step, (x, y) in enumerate(train_db):

            with tf.GradientTape() as tape:
                # [b, 32, 32, 3] => [b, 100]
                logits = model(x)
                # [b] => [b, 100]
                y_onehot = tf.one_hot(y, depth=100)
                # compute loss
                loss = tf.losses.categorical_crossentropy(y_onehot, logits, from_logits=True)
                loss = tf.reduce_mean(loss)

            grads = tape.gradient(loss, model.trainable_variables)
            optimizer.apply_gradients(zip(grads, model.trainable_variables))

            if step % 50 == 0:
                print(epoch, step, ‘loss:‘, float(loss))

        total_num = 0
        total_correct = 0
        for x, y in test_db:
            logits = model(x)
            prob = tf.nn.softmax(logits, axis=1)
            pred = tf.argmax(prob, axis=1)
            pred = tf.cast(pred, dtype=tf.int32)

            correct = tf.cast(tf.equal(pred, y), dtype=tf.int32)
            correct = tf.reduce_sum(correct)

            total_num += x.shape[0]
            total_correct += int(correct)

        acc = total_correct / total_num
        print(epoch, ‘acc:‘, acc)


if __name__ == ‘__main__‘:
    main()

(50000, 32, 32, 3) (50000,) (10000, 32, 32, 3) (10000,)
sample: (512, 32, 32, 3) (512,) tf.Tensor(-0.5, shape=(), dtype=float32) tf.Tensor(0.5, shape=(), dtype=float32)
Model: "res_net"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
sequential (Sequential)      multiple                  2048      
_________________________________________________________________
sequential_1 (Sequential)    multiple                  148736    
_________________________________________________________________
sequential_2 (Sequential)    multiple                  526976    
_________________________________________________________________
sequential_4 (Sequential)    multiple                  2102528   
_________________________________________________________________
sequential_6 (Sequential)    multiple                  8399360   
_________________________________________________________________
global_average_pooling2d (Gl multiple                  0         
_________________________________________________________________
dense (Dense)                multiple                  51300     
=================================================================
Total params: 11,230,948
Trainable params: 11,223,140
Non-trainable params: 7,808
_________________________________________________________________


WARNING: Logging before flag parsing goes to stderr.
W0601 16:59:57.619546 4664264128 optimizer_v2.py:928] Gradients does not exist for variables [‘sequential_2/basic_block_2/sequential_3/conv2d_7/kernel:0‘, ‘sequential_2/basic_block_2/sequential_3/conv2d_7/bias:0‘, ‘sequential_4/basic_block_4/sequential_5/conv2d_12/kernel:0‘, ‘sequential_4/basic_block_4/sequential_5/conv2d_12/bias:0‘, ‘sequential_6/basic_block_6/sequential_7/conv2d_17/kernel:0‘, ‘sequential_6/basic_block_6/sequential_7/conv2d_17/bias:0‘] when minimizing the loss.


0 0 loss: 4.60512638092041

Out of memory

• 1. decrease batch size
• 2. tune resnet[2,2,2,2]
• 3. try Google CoLab
• 4. buy new NVIDIA GPU Card