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
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
1. Train samples: (50000, 60, 60, 1) 2. Validation samples: (10000, 60, 60, 1) 3. Test samples: (10000, 60, 60, 1) 4. Input shape: (60, 60, 1)
T1、CNN_Init start
输出结果
1. T1、CNN_Init start! 2. _________________________________________________________________ 3. Layer (type) Output Shape Param # 4. ================================================================= 5. conv2d_1 (Conv2D) (None, 58, 58, 32) 320 6. _________________________________________________________________ 7. conv2d_2 (Conv2D) (None, 56, 56, 64) 18496 8. _________________________________________________________________ 9. max_pooling2d_1 (MaxPooling2 (None, 28, 28, 64) 0 10. _________________________________________________________________ 11. conv2d_3 (Conv2D) (None, 26, 26, 64) 36928 12. _________________________________________________________________ 13. max_pooling2d_2 (MaxPooling2 (None, 13, 13, 64) 0 14. _________________________________________________________________ 15. conv2d_4 (Conv2D) (None, 11, 11, 64) 36928 16. _________________________________________________________________ 17. dropout_1 (Dropout) (None, 11, 11, 64) 0 18. _________________________________________________________________ 19. flatten_1 (Flatten) (None, 7744) 0 20. _________________________________________________________________ 21. dense_1 (Dense) (None, 128) 991360 22. _________________________________________________________________ 23. dropout_2 (Dropout) (None, 128) 0 24. _________________________________________________________________ 25. dense_2 (Dense) (None, 10) 1290 26. ================================================================= 27. Total params: 1,085,322 28. Trainable params: 1,085,322 29. Non-trainable params: 0 30. _________________________________________________________________ 31. None 32. Train on 50000 samples, validate on 10000 samples 33. Epoch 1/30
核心代码
1. #(1)、定义模型结构 2. model = Sequential() 3. model.add(Conv2D(32, kernel_size=(3, 3), 4. activation='relu', 5. input_shape=input_shape)) 6. model.add(Conv2D(64, (3, 3), activation='relu')) 7. model.add(MaxPooling2D(pool_size=(2, 2))) 8. model.add(Conv2D(64, kernel_size=(3, 3), 9. activation='relu')) 10. model.add(MaxPooling2D(pool_size=(2, 2))) 11. model.add(Conv2D(64, (3, 3), activation='relu')) 12. model.add(Dropout(0.25)) 13. model.add(Flatten()) 14. model.add(Dense(128, activation='relu')) 15. model.add(Dropout(0.5)) 16. model.add(Dense(nb_classes, activation='softmax'))
T2、ST_CNN start
1. _________________________________________________________________ 2. Layer (type) Output Shape Param # 3. ================================================================= 4. conv2d_1 (Conv2D) (None, 56, 56, 32) 832 5. _________________________________________________________________ 6. activation_1 (Activation) (None, 56, 56, 32) 0 7. _________________________________________________________________ 8. max_pooling2d_1 (MaxPooling2 (None, 28, 28, 32) 0 9. _________________________________________________________________ 10. conv2d_2 (Conv2D) (None, 24, 24, 64) 51264 11. _________________________________________________________________ 12. activation_2 (Activation) (None, 24, 24, 64) 0 13. _________________________________________________________________ 14. conv2d_3 (Conv2D) (None, 22, 22, 64) 36928 15. _________________________________________________________________ 16. activation_3 (Activation) (None, 22, 22, 64) 0 17. _________________________________________________________________ 18. max_pooling2d_2 (MaxPooling2 (None, 11, 11, 64) 0 19. _________________________________________________________________ 20. flatten_1 (Flatten) (None, 7744) 0 21. _________________________________________________________________ 22. dense_1 (Dense) (None, 50) 387250 23. _________________________________________________________________ 24. activation_4 (Activation) (None, 50) 0 25. _________________________________________________________________ 26. dense_2 (Dense) (None, 6) 306 27. ================================================================= 28. Total params: 476,580 29. Trainable params: 476,580 30. Non-trainable params: 0 31. _________________________________________________________________ 32. None 33. _________________________________________________________________ 34. Layer (type) Output Shape Param # 35. ================================================================= 36. spatial_transformer_1 (Spati (None, 30, 30, 1) 476580 37. _________________________________________________________________ 38. conv2d_4 (Conv2D) (None, 28, 28, 32) 320 39. _________________________________________________________________ 40. dropout_1 (Dropout) (None, 28, 28, 32) 0 41. _________________________________________________________________ 42. conv2d_5 (Conv2D) (None, 26, 26, 64) 18496 43. _________________________________________________________________ 44. dropout_2 (Dropout) (None, 26, 26, 64) 0 45. _________________________________________________________________ 46. max_pooling2d_3 (MaxPooling2 (None, 13, 13, 64) 0 47. _________________________________________________________________ 48. conv2d_6 (Conv2D) (None, 11, 11, 64) 36928 49. _________________________________________________________________ 50. dropout_3 (Dropout) (None, 11, 11, 64) 0 51. _________________________________________________________________ 52. max_pooling2d_4 (MaxPooling2 (None, 5, 5, 64) 0 53. _________________________________________________________________ 54. flatten_2 (Flatten) (None, 1600) 0 55. _________________________________________________________________ 56. dense_3 (Dense) (None, 256) 409856 57. _________________________________________________________________ 58. dropout_4 (Dropout) (None, 256) 0 59. _________________________________________________________________ 60. activation_5 (Activation) (None, 256) 0 61. _________________________________________________________________ 62. dense_4 (Dense) (None, 10) 2570 63. _________________________________________________________________ 64. activation_6 (Activation) (None, 10) 0 65. ================================================================= 66. Total params: 944,750 67. Trainable params: 944,750 68. Non-trainable params: 0 69. _________________________________________________________________ 70. None 71. Train on 50000 samples, validate on 10000 samples 72. Epoch 1/30 73. - 974s - loss: 2.0926 - categorical_accuracy: 0.2345 - val_loss: 1.6258 - val_categorical_accuracy: 0.5949 74. Epoch 2/30 75. - 1007s - loss: 1.0926 - categorical_accuracy: 0.6387 - val_loss: 0.7963 - val_categorical_accuracy: 0.8433 76. Epoch 3/30 77. - 844s - loss: 0.6038 - categorical_accuracy: 0.8118 - val_loss: 0.4906 - val_categorical_accuracy: 0.8977 78. Epoch 4/30 79. - 851s - loss: 0.4351 - categorical_accuracy: 0.8648 - val_loss: 0.3909 - val_categorical_accuracy: 0.9160 80. Epoch 5/30 81. - 864s - loss: 0.3483 - categorical_accuracy: 0.8914 - val_loss: 0.3046 - val_categorical_accuracy: 0.9367 82. Epoch 6/30 83. - 872s - loss: 0.3158 - categorical_accuracy: 0.9027 - val_loss: 0.2826 - val_categorical_accuracy: 0.9349 84. Epoch 7/30 85. - 861s - loss: 0.2772 - categorical_accuracy: 0.9136 - val_loss: 0.3244 - val_categorical_accuracy: 0.9243 86. Epoch 8/30 87. - 862s - loss: 0.2414 - categorical_accuracy: 0.9251 - val_loss: 0.2228 - val_categorical_accuracy: 0.9600 88. Epoch 9/30 89. - 858s - loss: 0.2278 - categorical_accuracy: 0.9287 - val_loss: 0.2305 - val_categorical_accuracy: 0.9556 90. Epoch 10/30 91. - 860s - loss: 0.2150 - categorical_accuracy: 0.9328 - val_loss: 0.2119 - val_categorical_accuracy: 0.9600 92. Epoch 11/30 93. - 862s - loss: 0.2130 - categorical_accuracy: 0.9334 - val_loss: 0.1949 - val_categorical_accuracy: 0.9583 94. Epoch 12/30 95. - 855s - loss: 0.1917 - categorical_accuracy: 0.9410 - val_loss: 0.1841 - val_categorical_accuracy: 0.9595 96. Epoch 13/30 97. - 857s - loss: 0.1891 - categorical_accuracy: 0.9414 - val_loss: 0.2455 - val_categorical_accuracy: 0.9613 98. Epoch 14/30 99. - 862s - loss: 0.1865 - categorical_accuracy: 0.9423 - val_loss: 0.2044 - val_categorical_accuracy: 0.9629 100. Epoch 15/30 101. - 863s - loss: 0.1789 - categorical_accuracy: 0.9446 - val_loss: 0.2147 - val_categorical_accuracy: 0.9647 102. Epoch 16/30 103. - 855s - loss: 0.1708 - categorical_accuracy: 0.9460 - val_loss: 0.1748 - val_categorical_accuracy: 0.9692 104. Epoch 17/30 105. - 859s - loss: 0.1615 - categorical_accuracy: 0.9509 - val_loss: 0.1870 - val_categorical_accuracy: 0.9707 106. Epoch 18/30 107. - 862s - loss: 0.1538 - categorical_accuracy: 0.9514 - val_loss: 0.1906 - val_categorical_accuracy: 0.9689 108. Epoch 19/30 109. - 866s - loss: 0.1494 - categorical_accuracy: 0.9537 - val_loss: 0.1596 - val_categorical_accuracy: 0.9728 110. Epoch 20/30 111. - 864s - loss: 0.1490 - categorical_accuracy: 0.9537 - val_loss: 0.1821 - val_categorical_accuracy: 0.9692 112. Epoch 21/30 113. - 860s - loss: 0.1517 - categorical_accuracy: 0.9524 - val_loss: 0.1579 - val_categorical_accuracy: 0.9701 114. Epoch 22/30 115. - 859s - loss: 0.1506 - categorical_accuracy: 0.9539 - val_loss: 0.1595 - val_categorical_accuracy: 0.9712 116. Epoch 23/30 117. - 859s - loss: 0.1407 - categorical_accuracy: 0.9567 - val_loss: 0.1590 - val_categorical_accuracy: 0.9712 118. Epoch 24/30 119. - 856s - loss: 0.1361 - categorical_accuracy: 0.9569 - val_loss: 0.2160 - val_categorical_accuracy: 0.9723 120. Epoch 25/30 121. - 856s - loss: 0.1348 - categorical_accuracy: 0.9583 - val_loss: 0.1678 - val_categorical_accuracy: 0.9741 122. Epoch 26/30 123. - 856s - loss: 0.1298 - categorical_accuracy: 0.9596 - val_loss: 0.1820 - val_categorical_accuracy: 0.9707 124. Epoch 27/30 125. - 856s - loss: 0.1317 - categorical_accuracy: 0.9597 - val_loss: 0.1998 - val_categorical_accuracy: 0.9738 126. Epoch 28/30 127. - 855s - loss: 0.1325 - categorical_accuracy: 0.9594 - val_loss: 0.1991 - val_categorical_accuracy: 0.9674 128. Epoch 29/30 129. - 856s - loss: 0.1230 - categorical_accuracy: 0.9621 - val_loss: 0.1848 - val_categorical_accuracy: 0.9720 130. Epoch 30/30 131. - 856s - loss: 0.1246 - categorical_accuracy: 0.9611 - val_loss: 0.1754 - val_categorical_accuracy: 0.9755 132. 133. 32/10000 [..............................] - ETA: 59s 134. 64/10000 [..............................] - ETA: 59s 135. 96/10000 [..............................] - ETA: 59s 136. 128/10000 [..............................] - ETA: 57s 137. 160/10000 [..............................] - ETA: 56s 138. 192/10000 [..............................] - ETA: 55s 139. 224/10000 [..............................] - ETA: 55s 140. 256/10000 [..............................] - ETA: 54s 141. 288/10000 [..............................] - ETA: 54s 142. 320/10000 [..............................] - ETA: 54s 143. 352/10000 [>.............................] - ETA: 53s 144. 384/10000 [>.............................] - ETA: 53s 145. 416/10000 [>.............................] - ETA: 53s 146. 448/10000 [>.............................] - ETA: 52s 147. 480/10000 [>.............................] - ETA: 52s 148. 512/10000 [>.............................] - ETA: 52s 149. 544/10000 [>.............................] - ETA: 52s 150. 576/10000 [>.............................] - ETA: 52s 151. 608/10000 [>.............................] - ETA: 52s 152. 640/10000 [>.............................] - ETA: 52s 153. 672/10000 [=>............................] - ETA: 51s 154. 704/10000 [=>............................] - ETA: 51s 155. 736/10000 [=>............................] - ETA: 51s 156. 768/10000 [=>............................] - ETA: 51s 157. 800/10000 [=>............................] - ETA: 51s 158. 832/10000 [=>............................] - ETA: 50s 159. 864/10000 [=>............................] - ETA: 50s 160. 896/10000 [=>............................] - ETA: 50s 161. 928/10000 [=>............................] - ETA: 50s 162. 960/10000 [=>............................] - ETA: 50s 163. 992/10000 [=>............................] - ETA: 49s 164. 1024/10000 [==>...........................] - ETA: 49s 165. 1056/10000 [==>...........................] - ETA: 49s 166. 1088/10000 [==>...........................] - ETA: 49s 167. 1120/10000 [==>...........................] - ETA: 49s 168. 1152/10000 [==>...........................] - ETA: 49s 169. 1184/10000 [==>...........................] - ETA: 49s 170. 1216/10000 [==>...........................] - ETA: 48s 171. 1248/10000 [==>...........................] - ETA: 48s 172. 1280/10000 [==>...........................] - ETA: 48s 173. 1312/10000 [==>...........................] - ETA: 48s 174. 1344/10000 [===>..........................] - ETA: 48s 175. 1376/10000 [===>..........................] - ETA: 47s 176. 1408/10000 [===>..........................] - ETA: 47s 177. 1440/10000 [===>..........................] - ETA: 47s 178. 1472/10000 [===>..........................] - ETA: 47s 179. 1504/10000 [===>..........................] - ETA: 47s 180. 1536/10000 [===>..........................] - ETA: 46s 181. 1568/10000 [===>..........................] - ETA: 46s 182. 1600/10000 [===>..........................] - ETA: 46s 183. 1632/10000 [===>..........................] - ETA: 46s 184. 1664/10000 [===>..........................] - ETA: 46s 185. 1696/10000 [====>.........................] - ETA: 46s 186. 1728/10000 [====>.........................] - ETA: 46s 187. 1760/10000 [====>.........................] - ETA: 46s 188. 1792/10000 [====>.........................] - ETA: 46s 189. 1824/10000 [====>.........................] - ETA: 45s 190. 1856/10000 [====>.........................] - ETA: 45s 191. 1888/10000 [====>.........................] - ETA: 45s 192. 1920/10000 [====>.........................] - ETA: 45s 193. 1952/10000 [====>.........................] - ETA: 45s 194. 1984/10000 [====>.........................] - ETA: 45s 195. 2016/10000 [=====>........................] - ETA: 44s 196. 2048/10000 [=====>........................] - ETA: 44s 197. 2080/10000 [=====>........................] - ETA: 44s 198. 2112/10000 [=====>........................] - ETA: 44s 199. 2144/10000 [=====>........................] - ETA: 44s 200. 2176/10000 [=====>........................] - ETA: 44s 201. 2208/10000 [=====>........................] - ETA: 44s 202. 2240/10000 [=====>........................] - ETA: 43s 203. 2272/10000 [=====>........................] - ETA: 43s 204. 2304/10000 [=====>........................] - ETA: 43s 205. 2336/10000 [======>.......................] - ETA: 43s 206. 2368/10000 [======>.......................] - ETA: 43s 207. 2400/10000 [======>.......................] - ETA: 43s 208. 2432/10000 [======>.......................] - ETA: 42s 209. 2464/10000 [======>.......................] - ETA: 42s 210. 2496/10000 [======>.......................] - ETA: 42s 211. 2528/10000 [======>.......................] - ETA: 42s 212. 2560/10000 [======>.......................] - ETA: 42s 213. 2592/10000 [======>.......................] - ETA: 41s 214. 2624/10000 [======>.......................] - ETA: 41s 215. 2656/10000 [======>.......................] - ETA: 41s 216. 2688/10000 [=======>......................] - ETA: 41s 217. 2720/10000 [=======>......................] - ETA: 41s 218. 2752/10000 [=======>......................] - ETA: 41s 219. 2784/10000 [=======>......................] - ETA: 40s 220. 2816/10000 [=======>......................] - ETA: 40s 221. 2848/10000 [=======>......................] - ETA: 40s 222. 2880/10000 [=======>......................] - ETA: 40s 223. 2912/10000 [=======>......................] - ETA: 40s 224. 2944/10000 [=======>......................] - ETA: 39s 225. 2976/10000 [=======>......................] - ETA: 39s 226. 3008/10000 [========>.....................] - ETA: 39s 227. 3040/10000 [========>.....................] - ETA: 39s 228. 3072/10000 [========>.....................] - ETA: 39s 229. 3104/10000 [========>.....................] - ETA: 39s 230. 3136/10000 [========>.....................] - ETA: 38s 231. 3168/10000 [========>.....................] - ETA: 38s 232. 3200/10000 [========>.....................] - ETA: 38s 233. 3232/10000 [========>.....................] - ETA: 38s 234. 3264/10000 [========>.....................] - ETA: 38s 235. 3296/10000 [========>.....................] - ETA: 37s 236. 3328/10000 [========>.....................] - ETA: 37s 237. 3360/10000 [=========>....................] - ETA: 37s 238. 3392/10000 [=========>....................] - ETA: 37s 239. 3424/10000 [=========>....................] - ETA: 37s 240. 3456/10000 [=========>....................] - ETA: 36s 241. 3488/10000 [=========>....................] - ETA: 36s 242. 3520/10000 [=========>....................] - ETA: 36s 243. 3552/10000 [=========>....................] - ETA: 36s 244. 3584/10000 [=========>....................] - ETA: 36s 245. 3616/10000 [=========>....................] - ETA: 36s 246. 3648/10000 [=========>....................] - ETA: 35s 247. 3680/10000 [==========>...................] - ETA: 35s 248. 3712/10000 [==========>...................] - ETA: 35s 249. 3744/10000 [==========>...................] - ETA: 35s 250. 3776/10000 [==========>...................] - ETA: 35s 251. 3808/10000 [==========>...................] - ETA: 34s 252. 3840/10000 [==========>...................] - ETA: 34s 253. 3872/10000 [==========>...................] - ETA: 34s 254. 3904/10000 [==========>...................] - ETA: 34s 255. 3936/10000 [==========>...................] - ETA: 34s 256. 3968/10000 [==========>...................] - ETA: 33s 257. 4000/10000 [===========>..................] - ETA: 33s 258. 4032/10000 [===========>..................] - ETA: 33s 259. 4064/10000 [===========>..................] - ETA: 33s 260. 4096/10000 [===========>..................] - ETA: 33s 261. 4128/10000 [===========>..................] - ETA: 33s 262. 4160/10000 [===========>..................] - ETA: 32s 263. 4192/10000 [===========>..................] - ETA: 32s 264. 4224/10000 [===========>..................] - ETA: 32s 265. 4256/10000 [===========>..................] - ETA: 32s 266. 4288/10000 [===========>..................] - ETA: 32s 267. 4320/10000 [===========>..................] - ETA: 31s 268. 4352/10000 [============>.................] - ETA: 31s 269. 4384/10000 [============>.................] - ETA: 31s 270. 4416/10000 [============>.................] - ETA: 31s 271. 4448/10000 [============>.................] - ETA: 31s 272. 4480/10000 [============>.................] - ETA: 31s 273. 4512/10000 [============>.................] - ETA: 30s 274. 4544/10000 [============>.................] - ETA: 30s 275. 4576/10000 [============>.................] - ETA: 30s 276. 4608/10000 [============>.................] - ETA: 30s 277. 4640/10000 [============>.................] - ETA: 30s 278. 4672/10000 [=============>................] - ETA: 29s 279. 4704/10000 [=============>................] - ETA: 29s 280. 4736/10000 [=============>................] - ETA: 29s 281. 4768/10000 [=============>................] - ETA: 29s 282. 4800/10000 [=============>................] - ETA: 29s 283. 4832/10000 [=============>................] - ETA: 29s 284. 4864/10000 [=============>................] - ETA: 28s 285. 4896/10000 [=============>................] - ETA: 28s 286. 4928/10000 [=============>................] - ETA: 28s 287. 4960/10000 [=============>................] - ETA: 28s 288. 4992/10000 [=============>................] - ETA: 28s 289. 5024/10000 [==============>...............] - ETA: 27s 290. 5056/10000 [==============>...............] - ETA: 27s 291. 5088/10000 [==============>...............] - ETA: 27s 292. 5120/10000 [==============>...............] - ETA: 27s 293. 5152/10000 [==============>...............] - ETA: 27s 294. 5184/10000 [==============>...............] - ETA: 27s 295. 5216/10000 [==============>...............] - ETA: 26s 296. 5248/10000 [==============>...............] - ETA: 26s 297. 5280/10000 [==============>...............] - ETA: 26s 298. 5312/10000 [==============>...............] - ETA: 26s 299. 5344/10000 [===============>..............] - ETA: 26s 300. 5376/10000 [===============>..............] - ETA: 25s 301. 5408/10000 [===============>..............] - ETA: 25s 302. 5440/10000 [===============>..............] - ETA: 25s 303. 5472/10000 [===============>..............] - ETA: 25s 304. 5504/10000 [===============>..............] - ETA: 25s 305. 5536/10000 [===============>..............] - ETA: 25s 306. 5568/10000 [===============>..............] - ETA: 24s 307. 5600/10000 [===============>..............] - ETA: 24s 308. 5632/10000 [===============>..............] - ETA: 24s 309. 5664/10000 [===============>..............] - ETA: 24s 310. 5696/10000 [================>.............] - ETA: 24s 311. 5728/10000 [================>.............] - ETA: 23s 312. 5760/10000 [================>.............] - ETA: 23s 313. 5792/10000 [================>.............] - ETA: 23s 314. 5824/10000 [================>.............] - ETA: 23s 315. 5856/10000 [================>.............] - ETA: 23s 316. 5888/10000 [================>.............] - ETA: 23s 317. 5920/10000 [================>.............] - ETA: 22s 318. 5952/10000 [================>.............] - ETA: 22s 319. 5984/10000 [================>.............] - ETA: 22s 320. 6016/10000 [=================>............] - ETA: 22s 321. 6048/10000 [=================>............] - ETA: 22s 322. 6080/10000 [=================>............] - ETA: 21s 323. 6112/10000 [=================>............] - ETA: 21s 324. 6144/10000 [=================>............] - ETA: 21s 325. 6176/10000 [=================>............] - ETA: 21s 326. 6208/10000 [=================>............] - ETA: 21s 327. 6240/10000 [=================>............] - ETA: 21s 328. 6272/10000 [=================>............] - ETA: 20s 329. 6304/10000 [=================>............] - ETA: 20s 330. 6336/10000 [==================>...........] - ETA: 20s 331. 6368/10000 [==================>...........] - ETA: 20s 332. 6400/10000 [==================>...........] - ETA: 20s 333. 6432/10000 [==================>...........] - ETA: 19s 334. 6464/10000 [==================>...........] - ETA: 19s 335. 6496/10000 [==================>...........] - ETA: 19s 336. 6528/10000 [==================>...........] - ETA: 19s 337. 6560/10000 [==================>...........] - ETA: 19s 338. 6592/10000 [==================>...........] - ETA: 19s 339. 6624/10000 [==================>...........] - ETA: 18s 340. 6656/10000 [==================>...........] - ETA: 18s 341. 6688/10000 [===================>..........] - ETA: 18s 342. 6720/10000 [===================>..........] - ETA: 18s 343. 6752/10000 [===================>..........] - ETA: 18s 344. 6784/10000 [===================>..........] - ETA: 17s 345. 6816/10000 [===================>..........] - ETA: 17s 346. 6848/10000 [===================>..........] - ETA: 17s 347. 6880/10000 [===================>..........] - ETA: 17s 348. 6912/10000 [===================>..........] - ETA: 17s 349. 6944/10000 [===================>..........] - ETA: 17s 350. 6976/10000 [===================>..........] - ETA: 16s 351. 7008/10000 [====================>.........] - ETA: 16s 352. 7040/10000 [====================>.........] - ETA: 16s 353. 7072/10000 [====================>.........] - ETA: 16s 354. 7104/10000 [====================>.........] - ETA: 16s 355. 7136/10000 [====================>.........] - ETA: 16s 356. 7168/10000 [====================>.........] - ETA: 15s 357. 7200/10000 [====================>.........] - ETA: 15s 358. 7232/10000 [====================>.........] - ETA: 15s 359. 7264/10000 [====================>.........] - ETA: 15s 360. 7296/10000 [====================>.........] - ETA: 15s 361. 7328/10000 [====================>.........] - ETA: 14s 362. 7360/10000 [=====================>........] - ETA: 14s 363. 7392/10000 [=====================>........] - ETA: 14s 364. 7424/10000 [=====================>........] - ETA: 14s 365. 7456/10000 [=====================>........] - ETA: 14s 366. 7488/10000 [=====================>........] - ETA: 14s 367. 7520/10000 [=====================>........] - ETA: 13s 368. 7552/10000 [=====================>........] - ETA: 13s 369. 7584/10000 [=====================>........] - ETA: 13s 370. 7616/10000 [=====================>........] - ETA: 13s 371. 7648/10000 [=====================>........] - ETA: 13s 372. 7680/10000 [======================>.......] - ETA: 13s 373. 7712/10000 [======================>.......] - ETA: 12s 374. 7744/10000 [======================>.......] - ETA: 12s 375. 7776/10000 [======================>.......] - ETA: 12s 376. 7808/10000 [======================>.......] - ETA: 12s 377. 7840/10000 [======================>.......] - ETA: 12s 378. 7872/10000 [======================>.......] - ETA: 11s 379. 7904/10000 [======================>.......] - ETA: 11s 380. 7936/10000 [======================>.......] - ETA: 11s 381. 7968/10000 [======================>.......] - ETA: 11s 382. 8000/10000 [=======================>......] - ETA: 11s 383. 8032/10000 [=======================>......] - ETA: 11s 384. 8064/10000 [=======================>......] - ETA: 10s 385. 8096/10000 [=======================>......] - ETA: 10s 386. 8128/10000 [=======================>......] - ETA: 10s 387. 8160/10000 [=======================>......] - ETA: 10s 388. 8192/10000 [=======================>......] - ETA: 10s 389. 8224/10000 [=======================>......] - ETA: 9s 390. 8256/10000 [=======================>......] - ETA: 9s 391. 8288/10000 [=======================>......] - ETA: 9s 392. 8320/10000 [=======================>......] - ETA: 9s 393. 8352/10000 [========================>.....] - ETA: 9s 394. 8384/10000 [========================>.....] - ETA: 9s 395. 8416/10000 [========================>.....] - ETA: 8s 396. 8448/10000 [========================>.....] - ETA: 8s 397. 8480/10000 [========================>.....] - ETA: 8s 398. 8512/10000 [========================>.....] - ETA: 8s 399. 8544/10000 [========================>.....] - ETA: 8s 400. 8576/10000 [========================>.....] - ETA: 7s 401. 8608/10000 [========================>.....] - ETA: 7s 402. 8640/10000 [========================>.....] - ETA: 7s 403. 8672/10000 [=========================>....] - ETA: 7s 404. 8704/10000 [=========================>....] - ETA: 7s 405. 8736/10000 [=========================>....] - ETA: 7s 406. 8768/10000 [=========================>....] - ETA: 6s 407. 8800/10000 [=========================>....] - ETA: 6s 408. 8832/10000 [=========================>....] - ETA: 6s 409. 8864/10000 [=========================>....] - ETA: 6s 410. 8896/10000 [=========================>....] - ETA: 6s 411. 8928/10000 [=========================>....] - ETA: 6s 412. 8960/10000 [=========================>....] - ETA: 5s 413. 8992/10000 [=========================>....] - ETA: 5s 414. 9024/10000 [==========================>...] - ETA: 5s 415. 9056/10000 [==========================>...] - ETA: 5s 416. 9088/10000 [==========================>...] - ETA: 5s 417. 9120/10000 [==========================>...] - ETA: 4s 418. 9152/10000 [==========================>...] - ETA: 4s 419. 9184/10000 [==========================>...] - ETA: 4s 420. 9216/10000 [==========================>...] - ETA: 4s 421. 9248/10000 [==========================>...] - ETA: 4s 422. 9280/10000 [==========================>...] - ETA: 4s 423. 9312/10000 [==========================>...] - ETA: 3s 424. 9344/10000 [===========================>..] - ETA: 3s 425. 9376/10000 [===========================>..] - ETA: 3s 426. 9408/10000 [===========================>..] - ETA: 3s 427. 9440/10000 [===========================>..] - ETA: 3s 428. 9472/10000 [===========================>..] - ETA: 2s 429. 9504/10000 [===========================>..] - ETA: 2s 430. 9536/10000 [===========================>..] - ETA: 2s 431. 9568/10000 [===========================>..] - ETA: 2s 432. 9600/10000 [===========================>..] - ETA: 2s 433. 9632/10000 [===========================>..] - ETA: 2s 434. 9664/10000 [===========================>..] - ETA: 1s 435. 9696/10000 [============================>.] - ETA: 1s 436. 9728/10000 [============================>.] - ETA: 1s 437. 9760/10000 [============================>.] - ETA: 1s 438. 9792/10000 [============================>.] - ETA: 1s 439. 9824/10000 [============================>.] - ETA: 0s 440. 9856/10000 [============================>.] - ETA: 0s 441. 9888/10000 [============================>.] - ETA: 0s 442. 9920/10000 [============================>.] - ETA: 0s 443. 9952/10000 [============================>.] - ETA: 0s 444. 9984/10000 [============================>.] - ETA: 0s 445. 10000/10000 [==============================] - 56s 6ms/step 446.
核心代码
1. #(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 2. # localization net. TODO: try more conv layers, and do max pooling on X- and Y-axes respectively 3. locnet = Sequential() 4. # locnet.add(MaxPooling2D(pool_size=(2,2), input_shape=input_shape)) 5. # locnet.add(Convolution2D(32, (5, 5))) 6. locnet.add(Convolution2D(32, (5, 5), input_shape=input_shape)) 7. locnet.add(Activation('relu')) 8. # locnet.add(Dropout(0.2)) # 0.2 9. locnet.add(MaxPooling2D(pool_size=(2,2))) 10. locnet.add(Convolution2D(64, (5, 5))) 11. locnet.add(Activation('relu')) 12. # locnet.add(Dropout(0.2)) # 0.3 13. locnet.add(Convolution2D(64, (3, 3))) 14. locnet.add(Activation('relu')) 15. locnet.add(MaxPooling2D(pool_size=(2,2))) 16. 17. locnet.add(Flatten()) 18. locnet.add(Dense(50)) 19. locnet.add(Activation('relu')) 20. locnet.add(Dense(6, weights=weights)) 21. print(locnet.summary()) 22. 23. 24. #(3)、建立CNN网络 25. model = Sequential() 26. model.add(SpatialTransformer(localization_net=locnet, 27. output_size=(30,30), input_shape=input_shape)) 28. # model.add(Convolution2D(32, (3, 3), padding='same')) 29. # model.add(Activation('relu')) 30. # model.add(MaxPooling2D(pool_size=(2, 2))) 31. # model.add(Convolution2D(64, (3, 3))) 32. # model.add(Activation('relu')) 33. # model.add(MaxPooling2D(pool_size=(2, 2))) 34. # model.add(Dropout(0.5)) # 0.25 35. 36. # E: removed first 3 dropout layers 37. model.add(Conv2D(32, kernel_size=(3, 3), activation='relu')) 38. model.add(Dropout(0.5)) # 0.5 39. model.add(Conv2D(64, (3, 3), activation='relu')) 40. model.add(Dropout(0.5)) # 0.5 41. model.add(MaxPooling2D(pool_size=(2, 2))) 42. model.add(Conv2D(64, kernel_size=(3, 3), 43. activation='relu')) 44. model.add(Dropout(0.5)) # 0.5 45. model.add(MaxPooling2D(pool_size=(2, 2))) 46. # model.add(Conv2D(64, (3, 3), activation='relu')) 47. # model.add(Dropout(0.5)) 48. model.add(Flatten()) 49. model.add(Dense(256)) # 256 50. model.add(Dropout(0.5)) # 0.5 51. model.add(Activation('relu')) 52. model.add(Dense(nb_classes)) 53. model.add(Activation('softmax'))
