目录
基于tensorflow框架采用CNN(改进的AlexNet,训练/评估/推理)卷积神经网络算法实现猫狗图像分类识别
基于tensorflow框架采用CNN(改进的AlexNet,训练/评估/推理)卷积神经网络算法实现猫狗图像分类识别
数据集介绍
数据下载：Dogs vs. Cats Redux: Kernels Edition | Kaggle
train文件夹里有25000张狗和猫的图片。这个文件夹中的每个图像都有标签作为文件名的一部分。测试文件夹包含12500张图片，根据数字id命名。对于测试集中的每个图像，您应该预测图像是一只狗的概率(1 =狗，0 =猫)。
输出结果
使用model.ckpt-6000模型预测
预测错误的只有一个案例，如下所示
| 序号 | 使用model.ckpt-4000模型预测 | 使用model.ckpt-6000模型预测 | 使用model.ckpt-8000模型预测 | 使用model.ckpt-10000模型预测 | 使用model.ckpt-12000模型预测 | |
| 1 | cat | cat (1).jpg 猫的概率 0.631 | cat (1).jpg 狗的概率 0.740 | cat (1).jpg 狗的概率 0.781 | cat (1).jpg 狗的概率 0.976 | cat (1).jpg 狗的概率 0.991 |
| 2 | cat (10).jpg 狗的概率 0.697 | cat (10).jpg 猫的概率 0.566 | cat (10).jpg 猫的概率 0.925 | cat (10).jpg 猫的概率 0.925 | cat (10).jpg 猫的概率 0.816 | |
| 3 | cat (11).jpg 猫的概率 0.927 | cat (11).jpg 猫的概率 0.988 | cat (11).jpg 猫的概率 1.000 | cat (11).jpg 猫的概率 1.000 | cat (11).jpg 猫的概率 1.000 | |
| 4 | cat (12).jpg 狗的概率 0.746 | cat (12).jpg 狗的概率 0.723 | cat (12).jpg 狗的概率 0.822 | cat (12).jpg 狗的概率 0.998 | cat (12).jpg 狗的概率 1.000 | |
| 5 | cat (13).jpg 猫的概率 0.933 | cat (13).jpg 猫的概率 0.983 | cat (13).jpg 猫的概率 0.997 | cat (13).jpg 猫的概率 1.000 | cat (13).jpg 猫的概率 1.000 | |
| 6 | cat (14).jpg 狗的概率 0.657 | cat (14).jpg 猫的概率 0.597 | cat (14).jpg 狗的概率 0.758 | cat (14).jpg 狗的概率 0.695 | cat (14).jpg 猫的概率 0.544 | |
| 7 | cat (15).jpg 狗的概率 0.578 | cat (15).jpg 狗的概率 0.535 | cat (15).jpg 狗的概率 0.526 | cat (15).jpg 狗的概率 0.750 | cat (15).jpg 狗的概率 0.569 | |
| 8 | cat (2).jpg 猫的概率 0.649 | cat (2).jpg 猫的概率 0.637 | cat (2).jpg 猫的概率 0.844 | cat (2).jpg 猫的概率 0.996 | cat (2).jpg 猫的概率 0.998 | |
| 9 | cat (3).jpg 狗的概率 0.668 | cat (3).jpg 猫的概率 0.538 | cat (3).jpg 猫的概率 0.710 | cat (3).jpg 猫的概率 0.968 | cat (3).jpg 猫的概率 0.995 | |
| 10 | cat (4).jpg 狗的概率 0.856 | cat (4).jpg 狗的概率 0.780 | cat (4).jpg 狗的概率 0.831 | cat (4).jpg 狗的概率 0.974 | cat (4).jpg 狗的概率 0.976 | |
| 11 | cat (5).jpg 猫的概率 0.812 | cat (5).jpg 猫的概率 0.776 | cat (5).jpg 猫的概率 0.505 | cat (5).jpg 猫的概率 0.732 | cat (5).jpg 狗的概率 0.608 | |
| 12 | cat (6).jpg 猫的概率 0.524 | cat (6).jpg 狗的概率 0.661 | cat (6).jpg 狗的概率 0.748 | cat (6).jpg 狗的概率 0.970 | cat (6).jpg 狗的概率 0.987 | |
| 13 | cat (7).jpg 狗的概率 0.612 | cat (7).jpg 猫的概率 0.845 | cat (7).jpg 猫的概率 0.894 | cat (7).jpg 猫的概率 0.987 | cat (7).jpg 猫的概率 0.728 | |
| 14 | cat (8).jpg 狗的概率 0.823 | cat (8).jpg 狗的概率 0.948 | cat (8).jpg 狗的概率 0.920 | cat (8).jpg 狗的概率 0.982 | cat (8).jpg 狗的概率 0.999 | |
| 15 | cat (9).jpg 猫的概率 0.697 | cat (9).jpg 猫的概率 0.704 | cat (9).jpg 狗的概率 0.819 | cat (9).jpg 猫的概率 0.930 | cat (9).jpg 狗的概率 0.718 | |
| 16 | dog | dog (1).jpg 狗的概率 0.987 | dog (1).jpg 狗的概率 0.995 | dog (1).jpg 狗的概率 0.999 | dog (1).jpg 狗的概率 1.000 | dog (1).jpg 狗的概率 1.000 |
| 17 | dog (10).jpg 狗的概率 0.628 | dog (10).jpg 猫的概率 0.629 | dog (10).jpg 猫的概率 0.994 | dog (10).jpg 猫的概率 1.000 | dog (10).jpg 猫的概率 1.000 | |
| 18 | dog (11).jpg 狗的概率 0.804 | dog (11).jpg 狗的概率 0.879 | dog (11).jpg 狗的概率 0.993 | dog (11).jpg 狗的概率 1.000 | dog (11).jpg 狗的概率 1.000 | |
| 19 | dog (12).jpg 猫的概率 0.704 | dog (12).jpg 猫的概率 0.758 | dog (12).jpg 狗的概率 0.503 | dog (12).jpg 狗的概率 0.653 | dog (12).jpg 猫的概率 0.985 | |
| 20 | dog (13).jpg 狗的概率 0.987 | dog (13).jpg 狗的概率 0.997 | dog (13).jpg 狗的概率 1.000 | dog (13).jpg 狗的概率 1.000 | dog (13).jpg 狗的概率 1.000 | |
| 21 | dog (14).jpg 狗的概率 0.815 | dog (14).jpg 狗的概率 0.844 | dog (14).jpg 狗的概率 0.904 | dog (14).jpg 狗的概率 0.996 | dog (14).jpg 狗的概率 0.950 | |
| 22 | dog (15).jpg 狗的概率 0.917 | dog (15).jpg 狗的概率 0.984 | dog (15).jpg 狗的概率 0.999 | dog (15).jpg 狗的概率 1.000 | dog (15).jpg 狗的概率 1.000 | |
| 23 | dog (16).jpg 狗的概率 0.883 | dog (16).jpg 狗的概率 0.931 | dog (16).jpg 狗的概率 0.830 | dog (16).jpg 狗的概率 0.975 | dog (16).jpg 狗的概率 0.983 | |
| 24 | dog (2).jpg 狗的概率 0.934 | dog (2).jpg 狗的概率 0.982 | dog (2).jpg 狗的概率 0.998 | dog (2).jpg 狗的概率 1.000 | dog (2).jpg 狗的概率 1.000 | |
| 25 | dog (3).jpg 狗的概率 0.993 | dog (3).jpg 狗的概率 1.000 | dog (3).jpg 狗的概率 1.000 | dog (3).jpg 狗的概率 1.000 | dog (3).jpg 狗的概率 1.000 | |
| 26 | dog (4).jpg 狗的概率 0.693 | dog (4).jpg 狗的概率 0.754 | dog (4).jpg 狗的概率 0.976 | dog (4).jpg 狗的概率 0.515 | dog (4).jpg 狗的概率 0.995 | |
| 27 | dog (5).jpg 狗的概率 0.916 | dog (5).jpg 狗的概率 0.976 | dog (5).jpg 狗的概率 0.993 | dog (5).jpg 狗的概率 0.998 | dog (5).jpg 狗的概率 1.000 | |
| 28 | dog (6).jpg 狗的概率 0.947 | dog (6).jpg 狗的概率 0.989 | dog (6).jpg 狗的概率 0.999 | dog (6).jpg 狗的概率 1.000 | dog (6).jpg 狗的概率 1.000 | |
| 29 | dog (7).jpg 猫的概率 0.526 | dog (7).jpg 猫的概率 0.685 | dog (7).jpg 猫的概率 0.961 | dog (7).jpg 猫的概率 1.000 | dog (7).jpg 猫的概率 1.000 | |
| 30 | dog (8).jpg 狗的概率 0.981 | dog (8).jpg 狗的概率 0.998 | dog (8).jpg 狗的概率 1.000 | dog (8).jpg 狗的概率 1.000 | dog (8).jpg 狗的概率 1.000 | |
| 31 | dog (9).jpg 狗的概率 0.899 | dog (9).jpg 狗的概率 0.983 | dog (9).jpg 狗的概率 0.999 | dog (9).jpg 狗的概率 1.000 | dog (9).jpg 狗的概率 1.000 |
训练结果
1. Step 0, train loss = 0.69, train accuracy = 78.12% 2. Step 50, train loss = 0.69, train accuracy = 43.75% 3. Step 100, train loss = 0.70, train accuracy = 46.88% 4. Step 150, train loss = 0.65, train accuracy = 75.00% 5. Step 200, train loss = 0.66, train accuracy = 59.38% 6. Step 250, train loss = 0.66, train accuracy = 62.50% 7. Step 300, train loss = 0.72, train accuracy = 40.62% 8. Step 350, train loss = 0.66, train accuracy = 62.50% 9. Step 400, train loss = 0.58, train accuracy = 68.75% 10. Step 450, train loss = 0.70, train accuracy = 65.62% 11. Step 500, train loss = 0.68, train accuracy = 56.25% 12. Step 550, train loss = 0.51, train accuracy = 81.25% 13. Step 600, train loss = 0.54, train accuracy = 75.00% 14. Step 650, train loss = 0.64, train accuracy = 68.75% 15. Step 700, train loss = 0.69, train accuracy = 53.12% 16. Step 750, train loss = 0.57, train accuracy = 71.88% 17. Step 800, train loss = 0.80, train accuracy = 50.00% 18. Step 850, train loss = 0.62, train accuracy = 59.38% 19. Step 900, train loss = 0.59, train accuracy = 65.62% 20. Step 950, train loss = 0.54, train accuracy = 71.88% 21. Step 1000, train loss = 0.57, train accuracy = 68.75% 22. Step 1050, train loss = 0.56, train accuracy = 78.12% 23. Step 1100, train loss = 0.66, train accuracy = 59.38% 24. Step 1150, train loss = 0.50, train accuracy = 84.38% 25. Step 1200, train loss = 0.46, train accuracy = 81.25% 26. Step 1250, train loss = 0.57, train accuracy = 59.38% 27. Step 1300, train loss = 0.37, train accuracy = 81.25% 28. Step 1350, train loss = 0.64, train accuracy = 62.50% 29. Step 1400, train loss = 0.44, train accuracy = 81.25% 30. Step 1450, train loss = 0.46, train accuracy = 84.38% 31. Step 1500, train loss = 0.50, train accuracy = 71.88% 32. Step 1550, train loss = 0.58, train accuracy = 62.50% 33. Step 1600, train loss = 0.43, train accuracy = 75.00% 34. Step 1650, train loss = 0.55, train accuracy = 71.88% 35. Step 1700, train loss = 0.50, train accuracy = 71.88% 36. Step 1750, train loss = 0.46, train accuracy = 75.00% 37. Step 1800, train loss = 0.81, train accuracy = 53.12% 38. Step 1850, train loss = 0.41, train accuracy = 90.62% 39. Step 1900, train loss = 0.65, train accuracy = 68.75% 40. Step 1950, train loss = 0.37, train accuracy = 84.38% 41. Step 2000, train loss = 0.39, train accuracy = 81.25% 42. Step 2050, train loss = 0.45, train accuracy = 84.38% 43. Step 2100, train loss = 0.44, train accuracy = 78.12% 44. Step 2150, train loss = 0.59, train accuracy = 65.62% 45. Step 2200, train loss = 0.51, train accuracy = 78.12% 46. Step 2250, train loss = 0.42, train accuracy = 81.25% 47. Step 2300, train loss = 0.32, train accuracy = 87.50% 48. Step 2350, train loss = 0.48, train accuracy = 75.00% 49. Step 2400, train loss = 0.54, train accuracy = 71.88% 50. Step 2450, train loss = 0.51, train accuracy = 71.88% 51. Step 2500, train loss = 0.73, train accuracy = 59.38% 52. Step 2550, train loss = 0.52, train accuracy = 78.12% 53. Step 2600, train loss = 0.65, train accuracy = 62.50% 54. Step 2650, train loss = 0.52, train accuracy = 71.88% 55. Step 2700, train loss = 0.48, train accuracy = 71.88% 56. Step 2750, train loss = 0.37, train accuracy = 84.38% 57. Step 2800, train loss = 0.46, train accuracy = 78.12% 58. Step 2850, train loss = 0.40, train accuracy = 84.38% 59. Step 2900, train loss = 0.45, train accuracy = 81.25% 60. Step 2950, train loss = 0.36, train accuracy = 84.38% 61. Step 3000, train loss = 0.46, train accuracy = 75.00% 62. Step 3050, train loss = 0.53, train accuracy = 71.88% 63. Step 3100, train loss = 0.37, train accuracy = 84.38% 64. Step 3150, train loss = 0.53, train accuracy = 75.00% 65. Step 3200, train loss = 0.52, train accuracy = 75.00% 66. Step 3250, train loss = 0.62, train accuracy = 65.62% 67. Step 3300, train loss = 0.58, train accuracy = 71.88% 68. Step 3350, train loss = 0.71, train accuracy = 65.62% 69. Step 3400, train loss = 0.43, train accuracy = 78.12% 70. Step 3450, train loss = 0.46, train accuracy = 78.12% 71. Step 3500, train loss = 0.46, train accuracy = 71.88% 72. Step 3550, train loss = 0.53, train accuracy = 68.75% 73. Step 3600, train loss = 0.44, train accuracy = 75.00% 74. Step 3650, train loss = 0.55, train accuracy = 65.62% 75. Step 3700, train loss = 0.62, train accuracy = 75.00% 76. Step 3750, train loss = 0.48, train accuracy = 75.00% 77. Step 3800, train loss = 0.66, train accuracy = 53.12% 78. Step 3850, train loss = 0.53, train accuracy = 75.00% 79. Step 3900, train loss = 0.36, train accuracy = 81.25% 80. Step 3950, train loss = 0.37, train accuracy = 87.50% 81. Step 4000, train loss = 0.46, train accuracy = 78.12% 82. Step 4050, train loss = 0.36, train accuracy = 84.38% 83. Step 4100, train loss = 0.34, train accuracy = 78.12% 84. Step 4150, train loss = 0.48, train accuracy = 78.12% 85. Step 4200, train loss = 0.43, train accuracy = 87.50% 86. Step 4250, train loss = 0.34, train accuracy = 84.38% 87. Step 4300, train loss = 0.28, train accuracy = 87.50% 88. Step 4350, train loss = 0.19, train accuracy = 96.88% 89. Step 4400, train loss = 0.46, train accuracy = 71.88% 90. Step 4450, train loss = 0.33, train accuracy = 84.38% 91. Step 4500, train loss = 0.55, train accuracy = 75.00% 92. Step 4550, train loss = 0.31, train accuracy = 93.75% 93. Step 4600, train loss = 0.30, train accuracy = 84.38% 94. Step 4650, train loss = 0.38, train accuracy = 84.38% 95. Step 4700, train loss = 0.36, train accuracy = 84.38% 96. Step 4750, train loss = 0.32, train accuracy = 87.50% 97. Step 4800, train loss = 0.36, train accuracy = 81.25% 98. Step 4850, train loss = 0.36, train accuracy = 87.50% 99. Step 4900, train loss = 0.49, train accuracy = 71.88% 100. Step 4950, train loss = 0.51, train accuracy = 68.75% 101. Step 5000, train loss = 0.59, train accuracy = 68.75% 102. Step 5050, train loss = 0.55, train accuracy = 75.00% 103. Step 5100, train loss = 0.71, train accuracy = 68.75% 104. Step 5150, train loss = 0.48, train accuracy = 71.88% 105. Step 5200, train loss = 0.39, train accuracy = 90.62% 106. Step 5250, train loss = 0.49, train accuracy = 81.25% 107. Step 5300, train loss = 0.36, train accuracy = 81.25% 108. Step 5350, train loss = 0.31, train accuracy = 90.62% 109. Step 5400, train loss = 0.39, train accuracy = 87.50% 110. Step 5450, train loss = 0.34, train accuracy = 78.12% 111. Step 5500, train loss = 0.29, train accuracy = 84.38% 112. Step 5550, train loss = 0.21, train accuracy = 93.75% 113. Step 5600, train loss = 0.41, train accuracy = 78.12% 114. Step 5650, train loss = 0.38, train accuracy = 84.38% 115. Step 5700, train loss = 0.27, train accuracy = 87.50% 116. Step 5750, train loss = 0.24, train accuracy = 90.62% 117. Step 5800, train loss = 0.17, train accuracy = 96.88% 118. Step 5850, train loss = 0.23, train accuracy = 93.75% 119. Step 5900, train loss = 0.37, train accuracy = 71.88% 120. Step 5950, train loss = 0.49, train accuracy = 71.88% 121. Step 6000, train loss = 0.43, train accuracy = 81.25% 122. Step 6050, train loss = 0.33, train accuracy = 87.50% 123. Step 6100, train loss = 0.46, train accuracy = 75.00% 124. Step 6150, train loss = 0.61, train accuracy = 81.25% 125. Step 6200, train loss = 0.34, train accuracy = 84.38% 126. Step 6250, train loss = 0.63, train accuracy = 71.88% 127. Step 6300, train loss = 0.21, train accuracy = 90.62% 128. Step 6350, train loss = 0.21, train accuracy = 90.62% 129. Step 6400, train loss = 0.27, train accuracy = 87.50% 130. Step 6450, train loss = 0.17, train accuracy = 87.50% 131. Step 6500, train loss = 0.34, train accuracy = 87.50% 132. Step 6550, train loss = 0.34, train accuracy = 87.50% 133. Step 6600, train loss = 0.32, train accuracy = 84.38% 134. Step 6650, train loss = 0.39, train accuracy = 84.38% 135. Step 6700, train loss = 0.38, train accuracy = 84.38% 136. Step 6750, train loss = 0.41, train accuracy = 84.38% 137. Step 6800, train loss = 0.49, train accuracy = 81.25% 138. Step 6850, train loss = 0.36, train accuracy = 84.38% 139. Step 6900, train loss = 0.20, train accuracy = 93.75% 140. Step 6950, train loss = 0.13, train accuracy = 93.75% 141. Step 7000, train loss = 0.31, train accuracy = 87.50% 142. Step 7050, train loss = 0.18, train accuracy = 93.75% 143. Step 7100, train loss = 0.23, train accuracy = 90.62% 144. Step 7150, train loss = 0.13, train accuracy = 96.88% 145. Step 7200, train loss = 0.14, train accuracy = 96.88% 146. Step 7250, train loss = 0.32, train accuracy = 84.38% 147. Step 7300, train loss = 0.18, train accuracy = 93.75% 148. Step 7350, train loss = 0.14, train accuracy = 100.00% 149. Step 7400, train loss = 0.60, train accuracy = 75.00% 150. Step 7450, train loss = 0.20, train accuracy = 93.75% 151. Step 7500, train loss = 0.13, train accuracy = 93.75% 152. Step 7550, train loss = 0.22, train accuracy = 90.62% 153. Step 7600, train loss = 0.13, train accuracy = 96.88% 154. Step 7650, train loss = 0.20, train accuracy = 93.75% 155. Step 7700, train loss = 0.24, train accuracy = 90.62% 156. Step 7750, train loss = 0.19, train accuracy = 93.75% 157. Step 7800, train loss = 0.16, train accuracy = 93.75% 158. Step 7850, train loss = 0.08, train accuracy = 100.00% 159. Step 7900, train loss = 0.10, train accuracy = 96.88% 160. Step 7950, train loss = 0.13, train accuracy = 93.75% 161. Step 8000, train loss = 0.18, train accuracy = 90.62% 162. Step 8050, train loss = 0.27, train accuracy = 93.75% 163. Step 8100, train loss = 0.04, train accuracy = 100.00% 164. Step 8150, train loss = 0.27, train accuracy = 87.50% 165. Step 8200, train loss = 0.06, train accuracy = 96.88% 166. Step 8250, train loss = 0.12, train accuracy = 100.00% 167. Step 8300, train loss = 0.28, train accuracy = 87.50% 168. Step 8350, train loss = 0.24, train accuracy = 90.62% 169. Step 8400, train loss = 0.16, train accuracy = 93.75% 170. Step 8450, train loss = 0.11, train accuracy = 93.75% 171. Step 8500, train loss = 0.13, train accuracy = 96.88% 172. Step 8550, train loss = 0.05, train accuracy = 100.00% 173. Step 8600, train loss = 0.10, train accuracy = 93.75% 174. Step 8650, train loss = 0.14, train accuracy = 100.00% 175. Step 8700, train loss = 0.21, train accuracy = 90.62% 176. Step 8750, train loss = 0.09, train accuracy = 96.88% 177. Step 8800, train loss = 0.11, train accuracy = 96.88% 178. Step 8850, train loss = 0.10, train accuracy = 96.88% 179. Step 8900, train loss = 0.12, train accuracy = 93.75% 180. Step 8950, train loss = 0.48, train accuracy = 81.25% 181. Step 9000, train loss = 0.07, train accuracy = 100.00% 182. Step 9050, train loss = 0.03, train accuracy = 100.00% 183. Step 9100, train loss = 0.10, train accuracy = 93.75% 184. Step 9150, train loss = 0.05, train accuracy = 96.88% 185. Step 9200, train loss = 0.04, train accuracy = 100.00% 186. Step 9250, train loss = 0.03, train accuracy = 100.00% 187. Step 9300, train loss = 0.04, train accuracy = 96.88% 188. Step 9350, train loss = 0.08, train accuracy = 100.00% 189. Step 9400, train loss = 0.05, train accuracy = 100.00% 190. Step 9450, train loss = 0.15, train accuracy = 90.62% 191. Step 9500, train loss = 0.03, train accuracy = 100.00% 192. Step 9550, train loss = 0.05, train accuracy = 100.00% 193. Step 9600, train loss = 0.15, train accuracy = 96.88% 194. Step 9650, train loss = 0.03, train accuracy = 100.00% 195. Step 9700, train loss = 0.02, train accuracy = 100.00% 196. Step 9750, train loss = 0.08, train accuracy = 96.88% 197. Step 9800, train loss = 0.04, train accuracy = 100.00% 198. Step 9850, train loss = 0.06, train accuracy = 96.88% 199. Step 9900, train loss = 0.03, train accuracy = 100.00% 200. Step 9950, train loss = 0.03, train accuracy = 100.00% 201. Step 10000, train loss = 0.11, train accuracy = 93.75% 202. Step 10050, train loss = 0.02, train accuracy = 100.00% 203. Step 10100, train loss = 0.01, train accuracy = 100.00% 204. Step 10150, train loss = 0.05, train accuracy = 96.88% 205. Step 10200, train loss = 0.07, train accuracy = 96.88% 206. Step 10250, train loss = 0.06, train accuracy = 96.88% 207. Step 10300, train loss = 0.03, train accuracy = 100.00% 208. Step 10350, train loss = 0.08, train accuracy = 96.88% 209. Step 10400, train loss = 0.05, train accuracy = 96.88% 210. Step 10450, train loss = 0.02, train accuracy = 100.00% 211. Step 10500, train loss = 0.22, train accuracy = 93.75% 212. Step 10550, train loss = 0.06, train accuracy = 100.00% 213. Step 10600, train loss = 0.02, train accuracy = 100.00% 214. Step 10650, train loss = 0.02, train accuracy = 100.00% 215. Step 10700, train loss = 0.03, train accuracy = 100.00% 216. Step 10750, train loss = 0.15, train accuracy = 96.88% 217. Step 10800, train loss = 0.05, train accuracy = 100.00% 218. Step 10850, train loss = 0.02, train accuracy = 100.00% 219. Step 10900, train loss = 0.04, train accuracy = 96.88% 220. Step 10950, train loss = 0.05, train accuracy = 96.88% 221. Step 11000, train loss = 0.02, train accuracy = 100.00% 222. Step 11050, train loss = 0.10, train accuracy = 96.88% 223. Step 11100, train loss = 0.08, train accuracy = 96.88% 224. Step 11150, train loss = 0.02, train accuracy = 100.00% 225. Step 11200, train loss = 0.01, train accuracy = 100.00% 226. Step 11250, train loss = 0.06, train accuracy = 96.88% 227. Step 11300, train loss = 0.18, train accuracy = 93.75% 228. Step 11350, train loss = 0.02, train accuracy = 100.00% 229. Step 11400, train loss = 0.04, train accuracy = 100.00% 230. Step 11450, train loss = 0.03, train accuracy = 100.00% 231. Step 11500, train loss = 0.01, train accuracy = 100.00% 232. Step 11550, train loss = 0.02, train accuracy = 100.00%
核心代码
1. weights = tf.get_variable('weights', 2. shape=[3, 3, 3, 16], 3. dtype=tf.float32, 4. initializer=tf.truncated_normal_initializer(stddev=0.1, dtype=tf.float32)) 5. biases = tf.get_variable('biases', 6. shape=[16], 7. dtype=tf.float32, 8. initializer=tf.constant_initializer(0.1)) 9. conv = tf.nn.conv2d(images, weights, strides=[1, 1, 1, 1], padding='SAME') 10. pre_activation = tf.nn.bias_add(conv, biases) 11. conv1 = tf.nn.relu(pre_activation, name=scope.name) 12. with tf.variable_scope('pooling1_lrn') as scope: 13. pool1 = tf.nn.max_pool(conv1, ksize=[1, 3, 3, 1], strides=[1, 2, 2, 1], padding='SAME', name='pooling1') 14. norm1 = tf.nn.lrn(pool1, depth_radius=4, bias=1.0, alpha=0.001 / 9.0, beta=0.75, name='norm1') 15. 16. with tf.variable_scope('conv2') as scope: 17. weights = tf.get_variable('weights', 18. shape=[3, 3, 16, 16], 19. dtype=tf.float32, 20. initializer=tf.truncated_normal_initializer(stddev=0.1, dtype=tf.float32)) 21. biases = tf.get_variable('biases', 22. shape=[16], 23. dtype=tf.float32, 24. initializer=tf.constant_initializer(0.1)) 25. conv = tf.nn.conv2d(norm1, weights, strides=[1, 1, 1, 1], padding='SAME') 26. pre_activation = tf.nn.bias_add(conv, biases) 27. conv2 = tf.nn.relu(pre_activation, name='conv2') 28. 29. 30. with tf.variable_scope('pooling2_lrn') as scope: 31. norm2 = tf.nn.lrn(conv2, depth_radius=4, bias=1.0, alpha=0.001 / 9.0, beta=0.75, name='norm2') 32. pool2 = tf.nn.max_pool(norm2, ksize=[1, 3, 3, 1], strides=[1, 1, 1, 1], padding='SAME', name='pooling2') 33. 34. with tf.variable_scope('local3') as scope: 35. reshape = tf.reshape(pool2, shape=[batch_size, -1]) 36. dim = reshape.get_shape()[1].value 37. weights = tf.get_variable('weights', 38. shape=[dim, 128], 39. dtype=tf.float32, 40. initializer=tf.truncated_normal_initializer(stddev=0.005, dtype=tf.float32)) 41. biases = tf.get_variable('biases', 42. shape=[128], 43. dtype=tf.float32, 44. initializer=tf.constant_initializer(0.1)) 45. local3 = tf.nn.relu(tf.matmul(reshape, weights) + biases, name=scope.name) 46. 47. # local4 48. with tf.variable_scope('local4') as scope: 49. weights = tf.get_variable('weights', 50. shape=[128, 128], 51. dtype=tf.float32, 52. initializer=tf.truncated_normal_initializer(stddev=0.005, dtype=tf.float32)) 53. biases = tf.get_variable('biases', 54. shape=[128], 55. dtype=tf.float32, 56. initializer=tf.constant_initializer(0.1)) 57. local4 = tf.nn.relu(tf.matmul(local3, weights) + biases, name='local4') 58. 59. 60. with tf.variable_scope('softmax_linear') as scope: 61. weights = tf.get_variable('softmax_linear', 62. shape=[128, n_classes], 63. dtype=tf.float32, 64. initializer=tf.truncated_normal_initializer(stddev=0.005, dtype=tf.float32)) 65. biases = tf.get_variable('biases', 66. shape=[n_classes], 67. dtype=tf.float32, 68. initializer=tf.constant_initializer(0.1)) 69. softmax_linear = tf.add(tf.matmul(local4, weights), biases, name='softmax_linear')
