DL之Attention：基于ClutteredMNIST手写数字图片数据集分别利用CNN_Init、ST_CNN算法(CNN+SpatialTransformer)实现多分类预测
目录
基于ClutteredMNIST手写数字图片数据集分别利用CNN_Init、ST_CNN算法(CNN+SpatialTransformer)实现多分类预测
扫描二维码关注公众号，回复：
12558567 查看本文章
相关文章
DL之Attention：基于ClutteredMNIST手写数字图片数据集分别利用CNN_Init、ST_CNN算法(CNN+SpatialTransformer)实现多分类预测
DL之Attention：基于ClutteredMNIST手写数字图片数据集分别利用CNN_Init、ST_CNN算法(CNN+SpatialTransformer)实现多分类预测实现
基于ClutteredMNIST手写数字图片数据集分别利用CNN_Init、ST_CNN算法(CNN+SpatialTransformer)实现多分类预测
数据特征工程
Train samples: (50000, 60, 60, 1)
Validation samples: (10000, 60, 60, 1)
Test samples: (10000, 60, 60, 1)
Input shape: (60, 60, 1)T1、CNN_Init start
输出结果
T1、CNN_Init start!
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_1 (Conv2D) (None, 58, 58, 32) 320
_________________________________________________________________
conv2d_2 (Conv2D) (None, 56, 56, 64) 18496
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 28, 28, 64) 0
_________________________________________________________________
conv2d_3 (Conv2D) (None, 26, 26, 64) 36928
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 13, 13, 64) 0
_________________________________________________________________
conv2d_4 (Conv2D) (None, 11, 11, 64) 36928
_________________________________________________________________
dropout_1 (Dropout) (None, 11, 11, 64) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 7744) 0
_________________________________________________________________
dense_1 (Dense) (None, 128) 991360
_________________________________________________________________
dropout_2 (Dropout) (None, 128) 0
_________________________________________________________________
dense_2 (Dense) (None, 10) 1290
=================================================================
Total params: 1,085,322
Trainable params: 1,085,322
Non-trainable params: 0
_________________________________________________________________
None
Train on 50000 samples, validate on 10000 samples
Epoch 1/30核心代码
#(1)、定义模型结构
model = Sequential()
model.add(Conv2D(32, kernel_size=(3, 3),
activation='relu',
input_shape=input_shape))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(64, kernel_size=(3, 3),
activation='relu'))
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(Dropout(0.25))
model.add(Flatten())
model.add(Dense(128, activation='relu'))
model.add(Dropout(0.5))
model.add(Dense(nb_classes, activation='softmax'))T2、ST_CNN start
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
conv2d_1 (Conv2D) (None, 56, 56, 32) 832
_________________________________________________________________
activation_1 (Activation) (None, 56, 56, 32) 0
_________________________________________________________________
max_pooling2d_1 (MaxPooling2 (None, 28, 28, 32) 0
_________________________________________________________________
conv2d_2 (Conv2D) (None, 24, 24, 64) 51264
_________________________________________________________________
activation_2 (Activation) (None, 24, 24, 64) 0
_________________________________________________________________
conv2d_3 (Conv2D) (None, 22, 22, 64) 36928
_________________________________________________________________
activation_3 (Activation) (None, 22, 22, 64) 0
_________________________________________________________________
max_pooling2d_2 (MaxPooling2 (None, 11, 11, 64) 0
_________________________________________________________________
flatten_1 (Flatten) (None, 7744) 0
_________________________________________________________________
dense_1 (Dense) (None, 50) 387250
_________________________________________________________________
activation_4 (Activation) (None, 50) 0
_________________________________________________________________
dense_2 (Dense) (None, 6) 306
=================================================================
Total params: 476,580
Trainable params: 476,580
Non-trainable params: 0
_________________________________________________________________
None
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
spatial_transformer_1 (Spati (None, 30, 30, 1) 476580
_________________________________________________________________
conv2d_4 (Conv2D) (None, 28, 28, 32) 320
_________________________________________________________________
dropout_1 (Dropout) (None, 28, 28, 32) 0
_________________________________________________________________
conv2d_5 (Conv2D) (None, 26, 26, 64) 18496
_________________________________________________________________
dropout_2 (Dropout) (None, 26, 26, 64) 0
_________________________________________________________________
max_pooling2d_3 (MaxPooling2 (None, 13, 13, 64) 0
_________________________________________________________________
conv2d_6 (Conv2D) (None, 11, 11, 64) 36928
_________________________________________________________________
dropout_3 (Dropout) (None, 11, 11, 64) 0
_________________________________________________________________
max_pooling2d_4 (MaxPooling2 (None, 5, 5, 64) 0
_________________________________________________________________
flatten_2 (Flatten) (None, 1600) 0
_________________________________________________________________
dense_3 (Dense) (None, 256) 409856
_________________________________________________________________
dropout_4 (Dropout) (None, 256) 0
_________________________________________________________________
activation_5 (Activation) (None, 256) 0
_________________________________________________________________
dense_4 (Dense) (None, 10) 2570
_________________________________________________________________
activation_6 (Activation) (None, 10) 0
=================================================================
Total params: 944,750
Trainable params: 944,750
Non-trainable params: 0
_________________________________________________________________
None
Train on 50000 samples, validate on 10000 samples
Epoch 1/30
- 974s - loss: 2.0926 - categorical_accuracy: 0.2345 - val_loss: 1.6258 - val_categorical_accuracy: 0.5949
Epoch 2/30
- 1007s - loss: 1.0926 - categorical_accuracy: 0.6387 - val_loss: 0.7963 - val_categorical_accuracy: 0.8433
Epoch 3/30
- 844s - loss: 0.6038 - categorical_accuracy: 0.8118 - val_loss: 0.4906 - val_categorical_accuracy: 0.8977
Epoch 4/30
- 851s - loss: 0.4351 - categorical_accuracy: 0.8648 - val_loss: 0.3909 - val_categorical_accuracy: 0.9160
Epoch 5/30
- 864s - loss: 0.3483 - categorical_accuracy: 0.8914 - val_loss: 0.3046 - val_categorical_accuracy: 0.9367
Epoch 6/30
- 872s - loss: 0.3158 - categorical_accuracy: 0.9027 - val_loss: 0.2826 - val_categorical_accuracy: 0.9349
Epoch 7/30
- 861s - loss: 0.2772 - categorical_accuracy: 0.9136 - val_loss: 0.3244 - val_categorical_accuracy: 0.9243
Epoch 8/30
- 862s - loss: 0.2414 - categorical_accuracy: 0.9251 - val_loss: 0.2228 - val_categorical_accuracy: 0.9600
Epoch 9/30
- 858s - loss: 0.2278 - categorical_accuracy: 0.9287 - val_loss: 0.2305 - val_categorical_accuracy: 0.9556
Epoch 10/30
- 860s - loss: 0.2150 - categorical_accuracy: 0.9328 - val_loss: 0.2119 - val_categorical_accuracy: 0.9600
Epoch 11/30
- 862s - loss: 0.2130 - categorical_accuracy: 0.9334 - val_loss: 0.1949 - val_categorical_accuracy: 0.9583
Epoch 12/30
- 855s - loss: 0.1917 - categorical_accuracy: 0.9410 - val_loss: 0.1841 - val_categorical_accuracy: 0.9595
Epoch 13/30
- 857s - loss: 0.1891 - categorical_accuracy: 0.9414 - val_loss: 0.2455 - val_categorical_accuracy: 0.9613
Epoch 14/30
- 862s - loss: 0.1865 - categorical_accuracy: 0.9423 - val_loss: 0.2044 - val_categorical_accuracy: 0.9629
Epoch 15/30
- 863s - loss: 0.1789 - categorical_accuracy: 0.9446 - val_loss: 0.2147 - val_categorical_accuracy: 0.9647
Epoch 16/30
- 855s - loss: 0.1708 - categorical_accuracy: 0.9460 - val_loss: 0.1748 - val_categorical_accuracy: 0.9692
Epoch 17/30
- 859s - loss: 0.1615 - categorical_accuracy: 0.9509 - val_loss: 0.1870 - val_categorical_accuracy: 0.9707
Epoch 18/30
- 862s - loss: 0.1538 - categorical_accuracy: 0.9514 - val_loss: 0.1906 - val_categorical_accuracy: 0.9689
Epoch 19/30
- 866s - loss: 0.1494 - categorical_accuracy: 0.9537 - val_loss: 0.1596 - val_categorical_accuracy: 0.9728
Epoch 20/30
- 864s - loss: 0.1490 - categorical_accuracy: 0.9537 - val_loss: 0.1821 - val_categorical_accuracy: 0.9692
Epoch 21/30
- 860s - loss: 0.1517 - categorical_accuracy: 0.9524 - val_loss: 0.1579 - val_categorical_accuracy: 0.9701
Epoch 22/30
- 859s - loss: 0.1506 - categorical_accuracy: 0.9539 - val_loss: 0.1595 - val_categorical_accuracy: 0.9712
Epoch 23/30
- 859s - loss: 0.1407 - categorical_accuracy: 0.9567 - val_loss: 0.1590 - val_categorical_accuracy: 0.9712
Epoch 24/30
- 856s - loss: 0.1361 - categorical_accuracy: 0.9569 - val_loss: 0.2160 - val_categorical_accuracy: 0.9723
Epoch 25/30
- 856s - loss: 0.1348 - categorical_accuracy: 0.9583 - val_loss: 0.1678 - val_categorical_accuracy: 0.9741
Epoch 26/30
- 856s - loss: 0.1298 - categorical_accuracy: 0.9596 - val_loss: 0.1820 - val_categorical_accuracy: 0.9707
Epoch 27/30
- 856s - loss: 0.1317 - categorical_accuracy: 0.9597 - val_loss: 0.1998 - val_categorical_accuracy: 0.9738
Epoch 28/30
- 855s - loss: 0.1325 - categorical_accuracy: 0.9594 - val_loss: 0.1991 - val_categorical_accuracy: 0.9674
Epoch 29/30
- 856s - loss: 0.1230 - categorical_accuracy: 0.9621 - val_loss: 0.1848 - val_categorical_accuracy: 0.9720
Epoch 30/30
- 856s - loss: 0.1246 - categorical_accuracy: 0.9611 - val_loss: 0.1754 - val_categorical_accuracy: 0.9755
32/10000 [..............................] - ETA: 59s
64/10000 [..............................] - ETA: 59s
96/10000 [..............................] - ETA: 59s
128/10000 [..............................] - ETA: 57s
160/10000 [..............................] - ETA: 56s
192/10000 [..............................] - ETA: 55s
224/10000 [..............................] - ETA: 55s
256/10000 [..............................] - ETA: 54s
288/10000 [..............................] - ETA: 54s
320/10000 [..............................] - ETA: 54s
352/10000 [>.............................] - ETA: 53s
384/10000 [>.............................] - ETA: 53s
416/10000 [>.............................] - ETA: 53s
448/10000 [>.............................] - ETA: 52s
480/10000 [>.............................] - ETA: 52s
512/10000 [>.............................] - ETA: 52s
544/10000 [>.............................] - ETA: 52s
576/10000 [>.............................] - ETA: 52s
608/10000 [>.............................] - ETA: 52s
640/10000 [>.............................] - ETA: 52s
672/10000 [=>............................] - ETA: 51s
704/10000 [=>............................] - ETA: 51s
736/10000 [=>............................] - ETA: 51s
768/10000 [=>............................] - ETA: 51s
800/10000 [=>............................] - ETA: 51s
832/10000 [=>............................] - ETA: 50s
864/10000 [=>............................] - ETA: 50s
896/10000 [=>............................] - ETA: 50s
928/10000 [=>............................] - ETA: 50s
960/10000 [=>............................] - ETA: 50s
992/10000 [=>............................] - ETA: 49s
1024/10000 [==>...........................] - ETA: 49s
1056/10000 [==>...........................] - ETA: 49s
1088/10000 [==>...........................] - ETA: 49s
1120/10000 [==>...........................] - ETA: 49s
1152/10000 [==>...........................] - ETA: 49s
1184/10000 [==>...........................] - ETA: 49s
1216/10000 [==>...........................] - ETA: 48s
1248/10000 [==>...........................] - ETA: 48s
1280/10000 [==>...........................] - ETA: 48s
1312/10000 [==>...........................] - ETA: 48s
1344/10000 [===>..........................] - ETA: 48s
1376/10000 [===>..........................] - ETA: 47s
1408/10000 [===>..........................] - ETA: 47s
1440/10000 [===>..........................] - ETA: 47s
1472/10000 [===>..........................] - ETA: 47s
1504/10000 [===>..........................] - ETA: 47s
1536/10000 [===>..........................] - ETA: 46s
1568/10000 [===>..........................] - ETA: 46s
1600/10000 [===>..........................] - ETA: 46s
1632/10000 [===>..........................] - ETA: 46s
1664/10000 [===>..........................] - ETA: 46s
1696/10000 [====>.........................] - ETA: 46s
1728/10000 [====>.........................] - ETA: 46s
1760/10000 [====>.........................] - ETA: 46s
1792/10000 [====>.........................] - ETA: 46s
1824/10000 [====>.........................] - ETA: 45s
1856/10000 [====>.........................] - ETA: 45s
1888/10000 [====>.........................] - ETA: 45s
1920/10000 [====>.........................] - ETA: 45s
1952/10000 [====>.........................] - ETA: 45s
1984/10000 [====>.........................] - ETA: 45s
2016/10000 [=====>........................] - ETA: 44s
2048/10000 [=====>........................] - ETA: 44s
2080/10000 [=====>........................] - ETA: 44s
2112/10000 [=====>........................] - ETA: 44s
2144/10000 [=====>........................] - ETA: 44s
2176/10000 [=====>........................] - ETA: 44s
2208/10000 [=====>........................] - ETA: 44s
2240/10000 [=====>........................] - ETA: 43s
2272/10000 [=====>........................] - ETA: 43s
2304/10000 [=====>........................] - ETA: 43s
2336/10000 [======>.......................] - ETA: 43s
2368/10000 [======>.......................] - ETA: 43s
2400/10000 [======>.......................] - ETA: 43s
2432/10000 [======>.......................] - ETA: 42s
2464/10000 [======>.......................] - ETA: 42s
2496/10000 [======>.......................] - ETA: 42s
2528/10000 [======>.......................] - ETA: 42s
2560/10000 [======>.......................] - ETA: 42s
2592/10000 [======>.......................] - ETA: 41s
2624/10000 [======>.......................] - ETA: 41s
2656/10000 [======>.......................] - ETA: 41s
2688/10000 [=======>......................] - ETA: 41s
2720/10000 [=======>......................] - ETA: 41s
2752/10000 [=======>......................] - ETA: 41s
2784/10000 [=======>......................] - ETA: 40s
2816/10000 [=======>......................] - ETA: 40s
2848/10000 [=======>......................] - ETA: 40s
2880/10000 [=======>......................] - ETA: 40s
2912/10000 [=======>......................] - ETA: 40s
2944/10000 [=======>......................] - ETA: 39s
2976/10000 [=======>......................] - ETA: 39s
3008/10000 [========>.....................] - ETA: 39s
3040/10000 [========>.....................] - ETA: 39s
3072/10000 [========>.....................] - ETA: 39s
3104/10000 [========>.....................] - ETA: 39s
3136/10000 [========>.....................] - ETA: 38s
3168/10000 [========>.....................] - ETA: 38s
3200/10000 [========>.....................] - ETA: 38s
3232/10000 [========>.....................] - ETA: 38s
3264/10000 [========>.....................] - ETA: 38s
3296/10000 [========>.....................] - ETA: 37s
3328/10000 [========>.....................] - ETA: 37s
3360/10000 [=========>....................] - ETA: 37s
3392/10000 [=========>....................] - ETA: 37s
3424/10000 [=========>....................] - ETA: 37s
3456/10000 [=========>....................] - ETA: 36s
3488/10000 [=========>....................] - ETA: 36s
3520/10000 [=========>....................] - ETA: 36s
3552/10000 [=========>....................] - ETA: 36s
3584/10000 [=========>....................] - ETA: 36s
3616/10000 [=========>....................] - ETA: 36s
3648/10000 [=========>....................] - ETA: 35s
3680/10000 [==========>...................] - ETA: 35s
3712/10000 [==========>...................] - ETA: 35s
3744/10000 [==========>...................] - ETA: 35s
3776/10000 [==========>...................] - ETA: 35s
3808/10000 [==========>...................] - ETA: 34s
3840/10000 [==========>...................] - ETA: 34s
3872/10000 [==========>...................] - ETA: 34s
3904/10000 [==========>...................] - ETA: 34s
3936/10000 [==========>...................] - ETA: 34s
3968/10000 [==========>...................] - ETA: 33s
4000/10000 [===========>..................] - ETA: 33s
4032/10000 [===========>..................] - ETA: 33s
4064/10000 [===========>..................] - ETA: 33s
4096/10000 [===========>..................] - ETA: 33s
4128/10000 [===========>..................] - ETA: 33s
4160/10000 [===========>..................] - ETA: 32s
4192/10000 [===========>..................] - ETA: 32s
4224/10000 [===========>..................] - ETA: 32s
4256/10000 [===========>..................] - ETA: 32s
4288/10000 [===========>..................] - ETA: 32s
4320/10000 [===========>..................] - ETA: 31s
4352/10000 [============>.................] - ETA: 31s
4384/10000 [============>.................] - ETA: 31s
4416/10000 [============>.................] - ETA: 31s
4448/10000 [============>.................] - ETA: 31s
4480/10000 [============>.................] - ETA: 31s
4512/10000 [============>.................] - ETA: 30s
4544/10000 [============>.................] - ETA: 30s
4576/10000 [============>.................] - ETA: 30s
4608/10000 [============>.................] - ETA: 30s
4640/10000 [============>.................] - ETA: 30s
4672/10000 [=============>................] - ETA: 29s
4704/10000 [=============>................] - ETA: 29s
4736/10000 [=============>................] - ETA: 29s
4768/10000 [=============>................] - ETA: 29s
4800/10000 [=============>................] - ETA: 29s
4832/10000 [=============>................] - ETA: 29s
4864/10000 [=============>................] - ETA: 28s
4896/10000 [=============>................] - ETA: 28s
4928/10000 [=============>................] - ETA: 28s
4960/10000 [=============>................] - ETA: 28s
4992/10000 [=============>................] - ETA: 28s
5024/10000 [==============>...............] - ETA: 27s
5056/10000 [==============>...............] - ETA: 27s
5088/10000 [==============>...............] - ETA: 27s
5120/10000 [==============>...............] - ETA: 27s
5152/10000 [==============>...............] - ETA: 27s
5184/10000 [==============>...............] - ETA: 27s
5216/10000 [==============>...............] - ETA: 26s
5248/10000 [==============>...............] - ETA: 26s
5280/10000 [==============>...............] - ETA: 26s
5312/10000 [==============>...............] - ETA: 26s
5344/10000 [===============>..............] - ETA: 26s
5376/10000 [===============>..............] - ETA: 25s
5408/10000 [===============>..............] - ETA: 25s
5440/10000 [===============>..............] - ETA: 25s
5472/10000 [===============>..............] - ETA: 25s
5504/10000 [===============>..............] - ETA: 25s
5536/10000 [===============>..............] - ETA: 25s
5568/10000 [===============>..............] - ETA: 24s
5600/10000 [===============>..............] - ETA: 24s
5632/10000 [===============>..............] - ETA: 24s
5664/10000 [===============>..............] - ETA: 24s
5696/10000 [================>.............] - ETA: 24s
5728/10000 [================>.............] - ETA: 23s
5760/10000 [================>.............] - ETA: 23s
5792/10000 [================>.............] - ETA: 23s
5824/10000 [================>.............] - ETA: 23s
5856/10000 [================>.............] - ETA: 23s
5888/10000 [================>.............] - ETA: 23s
5920/10000 [================>.............] - ETA: 22s
5952/10000 [================>.............] - ETA: 22s
5984/10000 [================>.............] - ETA: 22s
6016/10000 [=================>............] - ETA: 22s
6048/10000 [=================>............] - ETA: 22s
6080/10000 [=================>............] - ETA: 21s
6112/10000 [=================>............] - ETA: 21s
6144/10000 [=================>............] - ETA: 21s
6176/10000 [=================>............] - ETA: 21s
6208/10000 [=================>............] - ETA: 21s
6240/10000 [=================>............] - ETA: 21s
6272/10000 [=================>............] - ETA: 20s
6304/10000 [=================>............] - ETA: 20s
6336/10000 [==================>...........] - ETA: 20s
6368/10000 [==================>...........] - ETA: 20s
6400/10000 [==================>...........] - ETA: 20s
6432/10000 [==================>...........] - ETA: 19s
6464/10000 [==================>...........] - ETA: 19s
6496/10000 [==================>...........] - ETA: 19s
6528/10000 [==================>...........] - ETA: 19s
6560/10000 [==================>...........] - ETA: 19s
6592/10000 [==================>...........] - ETA: 19s
6624/10000 [==================>...........] - ETA: 18s
6656/10000 [==================>...........] - ETA: 18s
6688/10000 [===================>..........] - ETA: 18s
6720/10000 [===================>..........] - ETA: 18s
6752/10000 [===================>..........] - ETA: 18s
6784/10000 [===================>..........] - ETA: 17s
6816/10000 [===================>..........] - ETA: 17s
6848/10000 [===================>..........] - ETA: 17s
6880/10000 [===================>..........] - ETA: 17s
6912/10000 [===================>..........] - ETA: 17s
6944/10000 [===================>..........] - ETA: 17s
6976/10000 [===================>..........] - ETA: 16s
7008/10000 [====================>.........] - ETA: 16s
7040/10000 [====================>.........] - ETA: 16s
7072/10000 [====================>.........] - ETA: 16s
7104/10000 [====================>.........] - ETA: 16s
7136/10000 [====================>.........] - ETA: 16s
7168/10000 [====================>.........] - ETA: 15s
7200/10000 [====================>.........] - ETA: 15s
7232/10000 [====================>.........] - ETA: 15s
7264/10000 [====================>.........] - ETA: 15s
7296/10000 [====================>.........] - ETA: 15s
7328/10000 [====================>.........] - ETA: 14s
7360/10000 [=====================>........] - ETA: 14s
7392/10000 [=====================>........] - ETA: 14s
7424/10000 [=====================>........] - ETA: 14s
7456/10000 [=====================>........] - ETA: 14s
7488/10000 [=====================>........] - ETA: 14s
7520/10000 [=====================>........] - ETA: 13s
7552/10000 [=====================>........] - ETA: 13s
7584/10000 [=====================>........] - ETA: 13s
7616/10000 [=====================>........] - ETA: 13s
7648/10000 [=====================>........] - ETA: 13s
7680/10000 [======================>.......] - ETA: 13s
7712/10000 [======================>.......] - ETA: 12s
7744/10000 [======================>.......] - ETA: 12s
7776/10000 [======================>.......] - ETA: 12s
7808/10000 [======================>.......] - ETA: 12s
7840/10000 [======================>.......] - ETA: 12s
7872/10000 [======================>.......] - ETA: 11s
7904/10000 [======================>.......] - ETA: 11s
7936/10000 [======================>.......] - ETA: 11s
7968/10000 [======================>.......] - ETA: 11s
8000/10000 [=======================>......] - ETA: 11s
8032/10000 [=======================>......] - ETA: 11s
8064/10000 [=======================>......] - ETA: 10s
8096/10000 [=======================>......] - ETA: 10s
8128/10000 [=======================>......] - ETA: 10s
8160/10000 [=======================>......] - ETA: 10s
8192/10000 [=======================>......] - ETA: 10s
8224/10000 [=======================>......] - ETA: 9s
8256/10000 [=======================>......] - ETA: 9s
8288/10000 [=======================>......] - ETA: 9s
8320/10000 [=======================>......] - ETA: 9s
8352/10000 [========================>.....] - ETA: 9s
8384/10000 [========================>.....] - ETA: 9s
8416/10000 [========================>.....] - ETA: 8s
8448/10000 [========================>.....] - ETA: 8s
8480/10000 [========================>.....] - ETA: 8s
8512/10000 [========================>.....] - ETA: 8s
8544/10000 [========================>.....] - ETA: 8s
8576/10000 [========================>.....] - ETA: 7s
8608/10000 [========================>.....] - ETA: 7s
8640/10000 [========================>.....] - ETA: 7s
8672/10000 [=========================>....] - ETA: 7s
8704/10000 [=========================>....] - ETA: 7s
8736/10000 [=========================>....] - ETA: 7s
8768/10000 [=========================>....] - ETA: 6s
8800/10000 [=========================>....] - ETA: 6s
8832/10000 [=========================>....] - ETA: 6s
8864/10000 [=========================>....] - ETA: 6s
8896/10000 [=========================>....] - ETA: 6s
8928/10000 [=========================>....] - ETA: 6s
8960/10000 [=========================>....] - ETA: 5s
8992/10000 [=========================>....] - ETA: 5s
9024/10000 [==========================>...] - ETA: 5s
9056/10000 [==========================>...] - ETA: 5s
9088/10000 [==========================>...] - ETA: 5s
9120/10000 [==========================>...] - ETA: 4s
9152/10000 [==========================>...] - ETA: 4s
9184/10000 [==========================>...] - ETA: 4s
9216/10000 [==========================>...] - ETA: 4s
9248/10000 [==========================>...] - ETA: 4s
9280/10000 [==========================>...] - ETA: 4s
9312/10000 [==========================>...] - ETA: 3s
9344/10000 [===========================>..] - ETA: 3s
9376/10000 [===========================>..] - ETA: 3s
9408/10000 [===========================>..] - ETA: 3s
9440/10000 [===========================>..] - ETA: 3s
9472/10000 [===========================>..] - ETA: 2s
9504/10000 [===========================>..] - ETA: 2s
9536/10000 [===========================>..] - ETA: 2s
9568/10000 [===========================>..] - ETA: 2s
9600/10000 [===========================>..] - ETA: 2s
9632/10000 [===========================>..] - ETA: 2s
9664/10000 [===========================>..] - ETA: 1s
9696/10000 [============================>.] - ETA: 1s
9728/10000 [============================>.] - ETA: 1s
9760/10000 [============================>.] - ETA: 1s
9792/10000 [============================>.] - ETA: 1s
9824/10000 [============================>.] - ETA: 0s
9856/10000 [============================>.] - ETA: 0s
9888/10000 [============================>.] - ETA: 0s
9920/10000 [============================>.] - ETA: 0s
9952/10000 [============================>.] - ETA: 0s
9984/10000 [============================>.] - ETA: 0s
10000/10000 [==============================] - 56s 6ms/step
核心代码
#(2)、建立ST定位网络：尝试更多的conv层，并分别在X轴和y轴上做最大池化
# localization net. TODO: try more conv layers, and do max pooling on X- and Y-axes respectively
locnet = Sequential()
# locnet.add(MaxPooling2D(pool_size=(2,2), input_shape=input_shape))
# locnet.add(Convolution2D(32, (5, 5)))
locnet.add(Convolution2D(32, (5, 5), input_shape=input_shape))
locnet.add(Activation('relu'))
# locnet.add(Dropout(0.2)) # 0.2
locnet.add(MaxPooling2D(pool_size=(2,2)))
locnet.add(Convolution2D(64, (5, 5)))
locnet.add(Activation('relu'))
# locnet.add(Dropout(0.2)) # 0.3
locnet.add(Convolution2D(64, (3, 3)))
locnet.add(Activation('relu'))
locnet.add(MaxPooling2D(pool_size=(2,2)))
locnet.add(Flatten())
locnet.add(Dense(50))
locnet.add(Activation('relu'))
locnet.add(Dense(6, weights=weights))
print(locnet.summary())
#(3)、建立CNN网络
model = Sequential()
model.add(SpatialTransformer(localization_net=locnet,
output_size=(30,30), input_shape=input_shape))
# model.add(Convolution2D(32, (3, 3), padding='same'))
# model.add(Activation('relu'))
# model.add(MaxPooling2D(pool_size=(2, 2)))
# model.add(Convolution2D(64, (3, 3)))
# model.add(Activation('relu'))
# model.add(MaxPooling2D(pool_size=(2, 2)))
# model.add(Dropout(0.5)) # 0.25
# E: removed first 3 dropout layers
model.add(Conv2D(32, kernel_size=(3, 3), activation='relu'))
model.add(Dropout(0.5)) # 0.5
model.add(Conv2D(64, (3, 3), activation='relu'))
model.add(Dropout(0.5)) # 0.5
model.add(MaxPooling2D(pool_size=(2, 2)))
model.add(Conv2D(64, kernel_size=(3, 3),
activation='relu'))
model.add(Dropout(0.5)) # 0.5
model.add(MaxPooling2D(pool_size=(2, 2)))
# model.add(Conv2D(64, (3, 3), activation='relu'))
# model.add(Dropout(0.5))
model.add(Flatten())
model.add(Dense(256)) # 256
model.add(Dropout(0.5)) # 0.5
model.add(Activation('relu'))
model.add(Dense(nb_classes))
model.add(Activation('softmax'))