API tools faq
paste
Guest User

Untitled

a guest
Feb 1st, 2020
520
0
Never
Add comment
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
text 81.28 KB | None | 0 0
raw download clone embed print report
  1. graph torch-jit-export (
  2. %input.1[FLOAT, 1x3x256x128]
  3. ) initializers (
  4. %classifier.bias[FLOAT, 1041]
  5. %classifier.weight[FLOAT, 1041x512]
  6. %conv1.bn.bias[FLOAT, 64]
  7. %conv1.bn.weight[FLOAT, 64]
  8. %conv1.conv.weight[FLOAT, 64x3x7x7]
  9. %conv2.0.IN.bias[FLOAT, 256]
  10. %conv2.0.IN.weight[FLOAT, 256]
  11. %conv2.0.conv1.bn.bias[FLOAT, 64]
  12. %conv2.0.conv1.bn.num_batches_tracked[INT64, scalar]
  13. %conv2.0.conv1.bn.running_mean[FLOAT, 64]
  14. %conv2.0.conv1.bn.running_var[FLOAT, 64]
  15. %conv2.0.conv1.bn.weight[FLOAT, 64]
  16. %conv2.0.conv1.conv.weight[FLOAT, 64x64x1x1]
  17. %conv2.0.conv2.0.layers.0.bn.bias[FLOAT, 64]
  18. %conv2.0.conv2.0.layers.0.bn.num_batches_tracked[INT64, scalar]
  19. %conv2.0.conv2.0.layers.0.bn.running_mean[FLOAT, 64]
  20. %conv2.0.conv2.0.layers.0.bn.running_var[FLOAT, 64]
  21. %conv2.0.conv2.0.layers.0.bn.weight[FLOAT, 64]
  22. %conv2.0.conv2.0.layers.0.conv1.weight[FLOAT, 64x64x1x1]
  23. %conv2.0.conv2.0.layers.0.conv2.weight[FLOAT, 64x1x3x3]
  24. %conv2.0.conv2.1.layers.0.bn.bias[FLOAT, 64]
  25. %conv2.0.conv2.1.layers.0.bn.num_batches_tracked[INT64, scalar]
  26. %conv2.0.conv2.1.layers.0.bn.running_mean[FLOAT, 64]
  27. %conv2.0.conv2.1.layers.0.bn.running_var[FLOAT, 64]
  28. %conv2.0.conv2.1.layers.0.bn.weight[FLOAT, 64]
  29. %conv2.0.conv2.1.layers.0.conv1.weight[FLOAT, 64x64x1x1]
  30. %conv2.0.conv2.1.layers.0.conv2.weight[FLOAT, 64x1x3x3]
  31. %conv2.0.conv2.1.layers.1.bn.bias[FLOAT, 64]
  32. %conv2.0.conv2.1.layers.1.bn.num_batches_tracked[INT64, scalar]
  33. %conv2.0.conv2.1.layers.1.bn.running_mean[FLOAT, 64]
  34. %conv2.0.conv2.1.layers.1.bn.running_var[FLOAT, 64]
  35. %conv2.0.conv2.1.layers.1.bn.weight[FLOAT, 64]
  36. %conv2.0.conv2.1.layers.1.conv1.weight[FLOAT, 64x64x1x1]
  37. %conv2.0.conv2.1.layers.1.conv2.weight[FLOAT, 64x1x3x3]
  38. %conv2.0.conv2.2.layers.0.bn.bias[FLOAT, 64]
  39. %conv2.0.conv2.2.layers.0.bn.num_batches_tracked[INT64, scalar]
  40. %conv2.0.conv2.2.layers.0.bn.running_mean[FLOAT, 64]
  41. %conv2.0.conv2.2.layers.0.bn.running_var[FLOAT, 64]
  42. %conv2.0.conv2.2.layers.0.bn.weight[FLOAT, 64]
  43. %conv2.0.conv2.2.layers.0.conv1.weight[FLOAT, 64x64x1x1]
  44. %conv2.0.conv2.2.layers.0.conv2.weight[FLOAT, 64x1x3x3]
  45. %conv2.0.conv2.2.layers.1.bn.bias[FLOAT, 64]
  46. %conv2.0.conv2.2.layers.1.bn.num_batches_tracked[INT64, scalar]
  47. %conv2.0.conv2.2.layers.1.bn.running_mean[FLOAT, 64]
  48. %conv2.0.conv2.2.layers.1.bn.running_var[FLOAT, 64]
  49. %conv2.0.conv2.2.layers.1.bn.weight[FLOAT, 64]
  50. %conv2.0.conv2.2.layers.1.conv1.weight[FLOAT, 64x64x1x1]
  51. %conv2.0.conv2.2.layers.1.conv2.weight[FLOAT, 64x1x3x3]
  52. %conv2.0.conv2.2.layers.2.bn.bias[FLOAT, 64]
  53. %conv2.0.conv2.2.layers.2.bn.num_batches_tracked[INT64, scalar]
  54. %conv2.0.conv2.2.layers.2.bn.running_mean[FLOAT, 64]
  55. %conv2.0.conv2.2.layers.2.bn.running_var[FLOAT, 64]
  56. %conv2.0.conv2.2.layers.2.bn.weight[FLOAT, 64]
  57. %conv2.0.conv2.2.layers.2.conv1.weight[FLOAT, 64x64x1x1]
  58. %conv2.0.conv2.2.layers.2.conv2.weight[FLOAT, 64x1x3x3]
  59. %conv2.0.conv2.3.layers.0.bn.bias[FLOAT, 64]
  60. %conv2.0.conv2.3.layers.0.bn.num_batches_tracked[INT64, scalar]
  61. %conv2.0.conv2.3.layers.0.bn.running_mean[FLOAT, 64]
  62. %conv2.0.conv2.3.layers.0.bn.running_var[FLOAT, 64]
  63. %conv2.0.conv2.3.layers.0.bn.weight[FLOAT, 64]
  64. %conv2.0.conv2.3.layers.0.conv1.weight[FLOAT, 64x64x1x1]
  65. %conv2.0.conv2.3.layers.0.conv2.weight[FLOAT, 64x1x3x3]
  66. %conv2.0.conv2.3.layers.1.bn.bias[FLOAT, 64]
  67. %conv2.0.conv2.3.layers.1.bn.num_batches_tracked[INT64, scalar]
  68. %conv2.0.conv2.3.layers.1.bn.running_mean[FLOAT, 64]
  69. %conv2.0.conv2.3.layers.1.bn.running_var[FLOAT, 64]
  70. %conv2.0.conv2.3.layers.1.bn.weight[FLOAT, 64]
  71. %conv2.0.conv2.3.layers.1.conv1.weight[FLOAT, 64x64x1x1]
  72. %conv2.0.conv2.3.layers.1.conv2.weight[FLOAT, 64x1x3x3]
  73. %conv2.0.conv2.3.layers.2.bn.bias[FLOAT, 64]
  74. %conv2.0.conv2.3.layers.2.bn.num_batches_tracked[INT64, scalar]
  75. %conv2.0.conv2.3.layers.2.bn.running_mean[FLOAT, 64]
  76. %conv2.0.conv2.3.layers.2.bn.running_var[FLOAT, 64]
  77. %conv2.0.conv2.3.layers.2.bn.weight[FLOAT, 64]
  78. %conv2.0.conv2.3.layers.2.conv1.weight[FLOAT, 64x64x1x1]
  79. %conv2.0.conv2.3.layers.2.conv2.weight[FLOAT, 64x1x3x3]
  80. %conv2.0.conv2.3.layers.3.bn.bias[FLOAT, 64]
  81. %conv2.0.conv2.3.layers.3.bn.num_batches_tracked[INT64, scalar]
  82. %conv2.0.conv2.3.layers.3.bn.running_mean[FLOAT, 64]
  83. %conv2.0.conv2.3.layers.3.bn.running_var[FLOAT, 64]
  84. %conv2.0.conv2.3.layers.3.bn.weight[FLOAT, 64]
  85. %conv2.0.conv2.3.layers.3.conv1.weight[FLOAT, 64x64x1x1]
  86. %conv2.0.conv2.3.layers.3.conv2.weight[FLOAT, 64x1x3x3]
  87. %conv2.0.conv3.conv.weight[FLOAT, 256x64x1x1]
  88. %conv2.0.downsample.bn.bias[FLOAT, 256]
  89. %conv2.0.downsample.bn.num_batches_tracked[INT64, scalar]
  90. %conv2.0.downsample.bn.running_mean[FLOAT, 256]
  91. %conv2.0.downsample.bn.running_var[FLOAT, 256]
  92. %conv2.0.downsample.bn.weight[FLOAT, 256]
  93. %conv2.0.downsample.conv.weight[FLOAT, 256x64x1x1]
  94. %conv2.0.gate.fc1.bias[FLOAT, 4]
  95. %conv2.0.gate.fc1.weight[FLOAT, 4x64x1x1]
  96. %conv2.0.gate.fc2.bias[FLOAT, 64]
  97. %conv2.0.gate.fc2.weight[FLOAT, 64x4x1x1]
  98. %conv2.1.IN.bias[FLOAT, 256]
  99. %conv2.1.IN.weight[FLOAT, 256]
  100. %conv2.1.conv1.bn.bias[FLOAT, 64]
  101. %conv2.1.conv1.bn.num_batches_tracked[INT64, scalar]
  102. %conv2.1.conv1.bn.running_mean[FLOAT, 64]
  103. %conv2.1.conv1.bn.running_var[FLOAT, 64]
  104. %conv2.1.conv1.bn.weight[FLOAT, 64]
  105. %conv2.1.conv1.conv.weight[FLOAT, 64x256x1x1]
  106. %conv2.1.conv2.0.layers.0.bn.bias[FLOAT, 64]
  107. %conv2.1.conv2.0.layers.0.bn.num_batches_tracked[INT64, scalar]
  108. %conv2.1.conv2.0.layers.0.bn.running_mean[FLOAT, 64]
  109. %conv2.1.conv2.0.layers.0.bn.running_var[FLOAT, 64]
  110. %conv2.1.conv2.0.layers.0.bn.weight[FLOAT, 64]
  111. %conv2.1.conv2.0.layers.0.conv1.weight[FLOAT, 64x64x1x1]
  112. %conv2.1.conv2.0.layers.0.conv2.weight[FLOAT, 64x1x3x3]
  113. %conv2.1.conv2.1.layers.0.bn.bias[FLOAT, 64]
  114. %conv2.1.conv2.1.layers.0.bn.num_batches_tracked[INT64, scalar]
  115. %conv2.1.conv2.1.layers.0.bn.running_mean[FLOAT, 64]
  116. %conv2.1.conv2.1.layers.0.bn.running_var[FLOAT, 64]
  117. %conv2.1.conv2.1.layers.0.bn.weight[FLOAT, 64]
  118. %conv2.1.conv2.1.layers.0.conv1.weight[FLOAT, 64x64x1x1]
  119. %conv2.1.conv2.1.layers.0.conv2.weight[FLOAT, 64x1x3x3]
  120. %conv2.1.conv2.1.layers.1.bn.bias[FLOAT, 64]
  121. %conv2.1.conv2.1.layers.1.bn.num_batches_tracked[INT64, scalar]
  122. %conv2.1.conv2.1.layers.1.bn.running_mean[FLOAT, 64]
  123. %conv2.1.conv2.1.layers.1.bn.running_var[FLOAT, 64]
  124. %conv2.1.conv2.1.layers.1.bn.weight[FLOAT, 64]
  125. %conv2.1.conv2.1.layers.1.conv1.weight[FLOAT, 64x64x1x1]
  126. %conv2.1.conv2.1.layers.1.conv2.weight[FLOAT, 64x1x3x3]
  127. %conv2.1.conv2.2.layers.0.bn.bias[FLOAT, 64]
  128. %conv2.1.conv2.2.layers.0.bn.num_batches_tracked[INT64, scalar]
  129. %conv2.1.conv2.2.layers.0.bn.running_mean[FLOAT, 64]
  130. %conv2.1.conv2.2.layers.0.bn.running_var[FLOAT, 64]
  131. %conv2.1.conv2.2.layers.0.bn.weight[FLOAT, 64]
  132. %conv2.1.conv2.2.layers.0.conv1.weight[FLOAT, 64x64x1x1]
  133. %conv2.1.conv2.2.layers.0.conv2.weight[FLOAT, 64x1x3x3]
  134. %conv2.1.conv2.2.layers.1.bn.bias[FLOAT, 64]
  135. %conv2.1.conv2.2.layers.1.bn.num_batches_tracked[INT64, scalar]
  136. %conv2.1.conv2.2.layers.1.bn.running_mean[FLOAT, 64]
  137. %conv2.1.conv2.2.layers.1.bn.running_var[FLOAT, 64]
  138. %conv2.1.conv2.2.layers.1.bn.weight[FLOAT, 64]
  139. %conv2.1.conv2.2.layers.1.conv1.weight[FLOAT, 64x64x1x1]
  140. %conv2.1.conv2.2.layers.1.conv2.weight[FLOAT, 64x1x3x3]
  141. %conv2.1.conv2.2.layers.2.bn.bias[FLOAT, 64]
  142. %conv2.1.conv2.2.layers.2.bn.num_batches_tracked[INT64, scalar]
  143. %conv2.1.conv2.2.layers.2.bn.running_mean[FLOAT, 64]
  144. %conv2.1.conv2.2.layers.2.bn.running_var[FLOAT, 64]
  145. %conv2.1.conv2.2.layers.2.bn.weight[FLOAT, 64]
  146. %conv2.1.conv2.2.layers.2.conv1.weight[FLOAT, 64x64x1x1]
  147. %conv2.1.conv2.2.layers.2.conv2.weight[FLOAT, 64x1x3x3]
  148. %conv2.1.conv2.3.layers.0.bn.bias[FLOAT, 64]
  149. %conv2.1.conv2.3.layers.0.bn.num_batches_tracked[INT64, scalar]
  150. %conv2.1.conv2.3.layers.0.bn.running_mean[FLOAT, 64]
  151. %conv2.1.conv2.3.layers.0.bn.running_var[FLOAT, 64]
  152. %conv2.1.conv2.3.layers.0.bn.weight[FLOAT, 64]
  153. %conv2.1.conv2.3.layers.0.conv1.weight[FLOAT, 64x64x1x1]
  154. %conv2.1.conv2.3.layers.0.conv2.weight[FLOAT, 64x1x3x3]
  155. %conv2.1.conv2.3.layers.1.bn.bias[FLOAT, 64]
  156. %conv2.1.conv2.3.layers.1.bn.num_batches_tracked[INT64, scalar]
  157. %conv2.1.conv2.3.layers.1.bn.running_mean[FLOAT, 64]
  158. %conv2.1.conv2.3.layers.1.bn.running_var[FLOAT, 64]
  159. %conv2.1.conv2.3.layers.1.bn.weight[FLOAT, 64]
  160. %conv2.1.conv2.3.layers.1.conv1.weight[FLOAT, 64x64x1x1]
  161. %conv2.1.conv2.3.layers.1.conv2.weight[FLOAT, 64x1x3x3]
  162. %conv2.1.conv2.3.layers.2.bn.bias[FLOAT, 64]
  163. %conv2.1.conv2.3.layers.2.bn.num_batches_tracked[INT64, scalar]
  164. %conv2.1.conv2.3.layers.2.bn.running_mean[FLOAT, 64]
  165. %conv2.1.conv2.3.layers.2.bn.running_var[FLOAT, 64]
  166. %conv2.1.conv2.3.layers.2.bn.weight[FLOAT, 64]
  167. %conv2.1.conv2.3.layers.2.conv1.weight[FLOAT, 64x64x1x1]
  168. %conv2.1.conv2.3.layers.2.conv2.weight[FLOAT, 64x1x3x3]
  169. %conv2.1.conv2.3.layers.3.bn.bias[FLOAT, 64]
  170. %conv2.1.conv2.3.layers.3.bn.num_batches_tracked[INT64, scalar]
  171. %conv2.1.conv2.3.layers.3.bn.running_mean[FLOAT, 64]
  172. %conv2.1.conv2.3.layers.3.bn.running_var[FLOAT, 64]
  173. %conv2.1.conv2.3.layers.3.bn.weight[FLOAT, 64]
  174. %conv2.1.conv2.3.layers.3.conv1.weight[FLOAT, 64x64x1x1]
  175. %conv2.1.conv2.3.layers.3.conv2.weight[FLOAT, 64x1x3x3]
  176. %conv2.1.conv3.conv.weight[FLOAT, 256x64x1x1]
  177. %conv2.1.gate.fc1.bias[FLOAT, 4]
  178. %conv2.1.gate.fc1.weight[FLOAT, 4x64x1x1]
  179. %conv2.1.gate.fc2.bias[FLOAT, 64]
  180. %conv2.1.gate.fc2.weight[FLOAT, 64x4x1x1]
  181. %conv3.0.conv1.bn.bias[FLOAT, 96]
  182. %conv3.0.conv1.bn.num_batches_tracked[INT64, scalar]
  183. %conv3.0.conv1.bn.running_mean[FLOAT, 96]
  184. %conv3.0.conv1.bn.running_var[FLOAT, 96]
  185. %conv3.0.conv1.bn.weight[FLOAT, 96]
  186. %conv3.0.conv1.conv.weight[FLOAT, 96x256x1x1]
  187. %conv3.0.conv2.0.layers.0.bn.bias[FLOAT, 96]
  188. %conv3.0.conv2.0.layers.0.bn.num_batches_tracked[INT64, scalar]
  189. %conv3.0.conv2.0.layers.0.bn.running_mean[FLOAT, 96]
  190. %conv3.0.conv2.0.layers.0.bn.running_var[FLOAT, 96]
  191. %conv3.0.conv2.0.layers.0.bn.weight[FLOAT, 96]
  192. %conv3.0.conv2.0.layers.0.conv1.weight[FLOAT, 96x96x1x1]
  193. %conv3.0.conv2.0.layers.0.conv2.weight[FLOAT, 96x1x3x3]
  194. %conv3.0.conv2.1.layers.0.bn.bias[FLOAT, 96]
  195. %conv3.0.conv2.1.layers.0.bn.num_batches_tracked[INT64, scalar]
  196. %conv3.0.conv2.1.layers.0.bn.running_mean[FLOAT, 96]
  197. %conv3.0.conv2.1.layers.0.bn.running_var[FLOAT, 96]
  198. %conv3.0.conv2.1.layers.0.bn.weight[FLOAT, 96]
  199. %conv3.0.conv2.1.layers.0.conv1.weight[FLOAT, 96x96x1x1]
  200. %conv3.0.conv2.1.layers.0.conv2.weight[FLOAT, 96x1x3x3]
  201. %conv3.0.conv2.1.layers.1.bn.bias[FLOAT, 96]
  202. %conv3.0.conv2.1.layers.1.bn.num_batches_tracked[INT64, scalar]
  203. %conv3.0.conv2.1.layers.1.bn.running_mean[FLOAT, 96]
  204. %conv3.0.conv2.1.layers.1.bn.running_var[FLOAT, 96]
  205. %conv3.0.conv2.1.layers.1.bn.weight[FLOAT, 96]
  206. %conv3.0.conv2.1.layers.1.conv1.weight[FLOAT, 96x96x1x1]
  207. %conv3.0.conv2.1.layers.1.conv2.weight[FLOAT, 96x1x3x3]
  208. %conv3.0.conv2.2.layers.0.bn.bias[FLOAT, 96]
  209. %conv3.0.conv2.2.layers.0.bn.num_batches_tracked[INT64, scalar]
  210. %conv3.0.conv2.2.layers.0.bn.running_mean[FLOAT, 96]
  211. %conv3.0.conv2.2.layers.0.bn.running_var[FLOAT, 96]
  212. %conv3.0.conv2.2.layers.0.bn.weight[FLOAT, 96]
  213. %conv3.0.conv2.2.layers.0.conv1.weight[FLOAT, 96x96x1x1]
  214. %conv3.0.conv2.2.layers.0.conv2.weight[FLOAT, 96x1x3x3]
  215. %conv3.0.conv2.2.layers.1.bn.bias[FLOAT, 96]
  216. %conv3.0.conv2.2.layers.1.bn.num_batches_tracked[INT64, scalar]
  217. %conv3.0.conv2.2.layers.1.bn.running_mean[FLOAT, 96]
  218. %conv3.0.conv2.2.layers.1.bn.running_var[FLOAT, 96]
  219. %conv3.0.conv2.2.layers.1.bn.weight[FLOAT, 96]
  220. %conv3.0.conv2.2.layers.1.conv1.weight[FLOAT, 96x96x1x1]
  221. %conv3.0.conv2.2.layers.1.conv2.weight[FLOAT, 96x1x3x3]
  222. %conv3.0.conv2.2.layers.2.bn.bias[FLOAT, 96]
  223. %conv3.0.conv2.2.layers.2.bn.num_batches_tracked[INT64, scalar]
  224. %conv3.0.conv2.2.layers.2.bn.running_mean[FLOAT, 96]
  225. %conv3.0.conv2.2.layers.2.bn.running_var[FLOAT, 96]
  226. %conv3.0.conv2.2.layers.2.bn.weight[FLOAT, 96]
  227. %conv3.0.conv2.2.layers.2.conv1.weight[FLOAT, 96x96x1x1]
  228. %conv3.0.conv2.2.layers.2.conv2.weight[FLOAT, 96x1x3x3]
  229. %conv3.0.conv2.3.layers.0.bn.bias[FLOAT, 96]
  230. %conv3.0.conv2.3.layers.0.bn.num_batches_tracked[INT64, scalar]
  231. %conv3.0.conv2.3.layers.0.bn.running_mean[FLOAT, 96]
  232. %conv3.0.conv2.3.layers.0.bn.running_var[FLOAT, 96]
  233. %conv3.0.conv2.3.layers.0.bn.weight[FLOAT, 96]
  234. %conv3.0.conv2.3.layers.0.conv1.weight[FLOAT, 96x96x1x1]
  235. %conv3.0.conv2.3.layers.0.conv2.weight[FLOAT, 96x1x3x3]
  236. %conv3.0.conv2.3.layers.1.bn.bias[FLOAT, 96]
  237. %conv3.0.conv2.3.layers.1.bn.num_batches_tracked[INT64, scalar]
  238. %conv3.0.conv2.3.layers.1.bn.running_mean[FLOAT, 96]
  239. %conv3.0.conv2.3.layers.1.bn.running_var[FLOAT, 96]
  240. %conv3.0.conv2.3.layers.1.bn.weight[FLOAT, 96]
  241. %conv3.0.conv2.3.layers.1.conv1.weight[FLOAT, 96x96x1x1]
  242. %conv3.0.conv2.3.layers.1.conv2.weight[FLOAT, 96x1x3x3]
  243. %conv3.0.conv2.3.layers.2.bn.bias[FLOAT, 96]
  244. %conv3.0.conv2.3.layers.2.bn.num_batches_tracked[INT64, scalar]
  245. %conv3.0.conv2.3.layers.2.bn.running_mean[FLOAT, 96]
  246. %conv3.0.conv2.3.layers.2.bn.running_var[FLOAT, 96]
  247. %conv3.0.conv2.3.layers.2.bn.weight[FLOAT, 96]
  248. %conv3.0.conv2.3.layers.2.conv1.weight[FLOAT, 96x96x1x1]
  249. %conv3.0.conv2.3.layers.2.conv2.weight[FLOAT, 96x1x3x3]
  250. %conv3.0.conv2.3.layers.3.bn.bias[FLOAT, 96]
  251. %conv3.0.conv2.3.layers.3.bn.num_batches_tracked[INT64, scalar]
  252. %conv3.0.conv2.3.layers.3.bn.running_mean[FLOAT, 96]
  253. %conv3.0.conv2.3.layers.3.bn.running_var[FLOAT, 96]
  254. %conv3.0.conv2.3.layers.3.bn.weight[FLOAT, 96]
  255. %conv3.0.conv2.3.layers.3.conv1.weight[FLOAT, 96x96x1x1]
  256. %conv3.0.conv2.3.layers.3.conv2.weight[FLOAT, 96x1x3x3]
  257. %conv3.0.conv3.bn.bias[FLOAT, 384]
  258. %conv3.0.conv3.bn.num_batches_tracked[INT64, scalar]
  259. %conv3.0.conv3.bn.running_mean[FLOAT, 384]
  260. %conv3.0.conv3.bn.running_var[FLOAT, 384]
  261. %conv3.0.conv3.bn.weight[FLOAT, 384]
  262. %conv3.0.conv3.conv.weight[FLOAT, 384x96x1x1]
  263. %conv3.0.downsample.bn.bias[FLOAT, 384]
  264. %conv3.0.downsample.bn.num_batches_tracked[INT64, scalar]
  265. %conv3.0.downsample.bn.running_mean[FLOAT, 384]
  266. %conv3.0.downsample.bn.running_var[FLOAT, 384]
  267. %conv3.0.downsample.bn.weight[FLOAT, 384]
  268. %conv3.0.downsample.conv.weight[FLOAT, 384x256x1x1]
  269. %conv3.0.gate.fc1.bias[FLOAT, 6]
  270. %conv3.0.gate.fc1.weight[FLOAT, 6x96x1x1]
  271. %conv3.0.gate.fc2.bias[FLOAT, 96]
  272. %conv3.0.gate.fc2.weight[FLOAT, 96x6x1x1]
  273. %conv3.1.IN.bias[FLOAT, 384]
  274. %conv3.1.IN.weight[FLOAT, 384]
  275. %conv3.1.conv1.bn.bias[FLOAT, 96]
  276. %conv3.1.conv1.bn.num_batches_tracked[INT64, scalar]
  277. %conv3.1.conv1.bn.running_mean[FLOAT, 96]
  278. %conv3.1.conv1.bn.running_var[FLOAT, 96]
  279. %conv3.1.conv1.bn.weight[FLOAT, 96]
  280. %conv3.1.conv1.conv.weight[FLOAT, 96x384x1x1]
  281. %conv3.1.conv2.0.layers.0.bn.bias[FLOAT, 96]
  282. %conv3.1.conv2.0.layers.0.bn.num_batches_tracked[INT64, scalar]
  283. %conv3.1.conv2.0.layers.0.bn.running_mean[FLOAT, 96]
  284. %conv3.1.conv2.0.layers.0.bn.running_var[FLOAT, 96]
  285. %conv3.1.conv2.0.layers.0.bn.weight[FLOAT, 96]
  286. %conv3.1.conv2.0.layers.0.conv1.weight[FLOAT, 96x96x1x1]
  287. %conv3.1.conv2.0.layers.0.conv2.weight[FLOAT, 96x1x3x3]
  288. %conv3.1.conv2.1.layers.0.bn.bias[FLOAT, 96]
  289. %conv3.1.conv2.1.layers.0.bn.num_batches_tracked[INT64, scalar]
  290. %conv3.1.conv2.1.layers.0.bn.running_mean[FLOAT, 96]
  291. %conv3.1.conv2.1.layers.0.bn.running_var[FLOAT, 96]
  292. %conv3.1.conv2.1.layers.0.bn.weight[FLOAT, 96]
  293. %conv3.1.conv2.1.layers.0.conv1.weight[FLOAT, 96x96x1x1]
  294. %conv3.1.conv2.1.layers.0.conv2.weight[FLOAT, 96x1x3x3]
  295. %conv3.1.conv2.1.layers.1.bn.bias[FLOAT, 96]
  296. %conv3.1.conv2.1.layers.1.bn.num_batches_tracked[INT64, scalar]
  297. %conv3.1.conv2.1.layers.1.bn.running_mean[FLOAT, 96]
  298. %conv3.1.conv2.1.layers.1.bn.running_var[FLOAT, 96]
  299. %conv3.1.conv2.1.layers.1.bn.weight[FLOAT, 96]
  300. %conv3.1.conv2.1.layers.1.conv1.weight[FLOAT, 96x96x1x1]
  301. %conv3.1.conv2.1.layers.1.conv2.weight[FLOAT, 96x1x3x3]
  302. %conv3.1.conv2.2.layers.0.bn.bias[FLOAT, 96]
  303. %conv3.1.conv2.2.layers.0.bn.num_batches_tracked[INT64, scalar]
  304. %conv3.1.conv2.2.layers.0.bn.running_mean[FLOAT, 96]
  305. %conv3.1.conv2.2.layers.0.bn.running_var[FLOAT, 96]
  306. %conv3.1.conv2.2.layers.0.bn.weight[FLOAT, 96]
  307. %conv3.1.conv2.2.layers.0.conv1.weight[FLOAT, 96x96x1x1]
  308. %conv3.1.conv2.2.layers.0.conv2.weight[FLOAT, 96x1x3x3]
  309. %conv3.1.conv2.2.layers.1.bn.bias[FLOAT, 96]
  310. %conv3.1.conv2.2.layers.1.bn.num_batches_tracked[INT64, scalar]
  311. %conv3.1.conv2.2.layers.1.bn.running_mean[FLOAT, 96]
  312. %conv3.1.conv2.2.layers.1.bn.running_var[FLOAT, 96]
  313. %conv3.1.conv2.2.layers.1.bn.weight[FLOAT, 96]
  314. %conv3.1.conv2.2.layers.1.conv1.weight[FLOAT, 96x96x1x1]
  315. %conv3.1.conv2.2.layers.1.conv2.weight[FLOAT, 96x1x3x3]
  316. %conv3.1.conv2.2.layers.2.bn.bias[FLOAT, 96]
  317. %conv3.1.conv2.2.layers.2.bn.num_batches_tracked[INT64, scalar]
  318. %conv3.1.conv2.2.layers.2.bn.running_mean[FLOAT, 96]
  319. %conv3.1.conv2.2.layers.2.bn.running_var[FLOAT, 96]
  320. %conv3.1.conv2.2.layers.2.bn.weight[FLOAT, 96]
  321. %conv3.1.conv2.2.layers.2.conv1.weight[FLOAT, 96x96x1x1]
  322. %conv3.1.conv2.2.layers.2.conv2.weight[FLOAT, 96x1x3x3]
  323. %conv3.1.conv2.3.layers.0.bn.bias[FLOAT, 96]
  324. %conv3.1.conv2.3.layers.0.bn.num_batches_tracked[INT64, scalar]
  325. %conv3.1.conv2.3.layers.0.bn.running_mean[FLOAT, 96]
  326. %conv3.1.conv2.3.layers.0.bn.running_var[FLOAT, 96]
  327. %conv3.1.conv2.3.layers.0.bn.weight[FLOAT, 96]
  328. %conv3.1.conv2.3.layers.0.conv1.weight[FLOAT, 96x96x1x1]
  329. %conv3.1.conv2.3.layers.0.conv2.weight[FLOAT, 96x1x3x3]
  330. %conv3.1.conv2.3.layers.1.bn.bias[FLOAT, 96]
  331. %conv3.1.conv2.3.layers.1.bn.num_batches_tracked[INT64, scalar]
  332. %conv3.1.conv2.3.layers.1.bn.running_mean[FLOAT, 96]
  333. %conv3.1.conv2.3.layers.1.bn.running_var[FLOAT, 96]
  334. %conv3.1.conv2.3.layers.1.bn.weight[FLOAT, 96]
  335. %conv3.1.conv2.3.layers.1.conv1.weight[FLOAT, 96x96x1x1]
  336. %conv3.1.conv2.3.layers.1.conv2.weight[FLOAT, 96x1x3x3]
  337. %conv3.1.conv2.3.layers.2.bn.bias[FLOAT, 96]
  338. %conv3.1.conv2.3.layers.2.bn.num_batches_tracked[INT64, scalar]
  339. %conv3.1.conv2.3.layers.2.bn.running_mean[FLOAT, 96]
  340. %conv3.1.conv2.3.layers.2.bn.running_var[FLOAT, 96]
  341. %conv3.1.conv2.3.layers.2.bn.weight[FLOAT, 96]
  342. %conv3.1.conv2.3.layers.2.conv1.weight[FLOAT, 96x96x1x1]
  343. %conv3.1.conv2.3.layers.2.conv2.weight[FLOAT, 96x1x3x3]
  344. %conv3.1.conv2.3.layers.3.bn.bias[FLOAT, 96]
  345. %conv3.1.conv2.3.layers.3.bn.num_batches_tracked[INT64, scalar]
  346. %conv3.1.conv2.3.layers.3.bn.running_mean[FLOAT, 96]
  347. %conv3.1.conv2.3.layers.3.bn.running_var[FLOAT, 96]
  348. %conv3.1.conv2.3.layers.3.bn.weight[FLOAT, 96]
  349. %conv3.1.conv2.3.layers.3.conv1.weight[FLOAT, 96x96x1x1]
  350. %conv3.1.conv2.3.layers.3.conv2.weight[FLOAT, 96x1x3x3]
  351. %conv3.1.conv3.conv.weight[FLOAT, 384x96x1x1]
  352. %conv3.1.gate.fc1.bias[FLOAT, 6]
  353. %conv3.1.gate.fc1.weight[FLOAT, 6x96x1x1]
  354. %conv3.1.gate.fc2.bias[FLOAT, 96]
  355. %conv3.1.gate.fc2.weight[FLOAT, 96x6x1x1]
  356. %conv4.0.IN.bias[FLOAT, 512]
  357. %conv4.0.IN.weight[FLOAT, 512]
  358. %conv4.0.conv1.bn.bias[FLOAT, 128]
  359. %conv4.0.conv1.bn.num_batches_tracked[INT64, scalar]
  360. %conv4.0.conv1.bn.running_mean[FLOAT, 128]
  361. %conv4.0.conv1.bn.running_var[FLOAT, 128]
  362. %conv4.0.conv1.bn.weight[FLOAT, 128]
  363. %conv4.0.conv1.conv.weight[FLOAT, 128x384x1x1]
  364. %conv4.0.conv2.0.layers.0.bn.bias[FLOAT, 128]
  365. %conv4.0.conv2.0.layers.0.bn.num_batches_tracked[INT64, scalar]
  366. %conv4.0.conv2.0.layers.0.bn.running_mean[FLOAT, 128]
  367. %conv4.0.conv2.0.layers.0.bn.running_var[FLOAT, 128]
  368. %conv4.0.conv2.0.layers.0.bn.weight[FLOAT, 128]
  369. %conv4.0.conv2.0.layers.0.conv1.weight[FLOAT, 128x128x1x1]
  370. %conv4.0.conv2.0.layers.0.conv2.weight[FLOAT, 128x1x3x3]
  371. %conv4.0.conv2.1.layers.0.bn.bias[FLOAT, 128]
  372. %conv4.0.conv2.1.layers.0.bn.num_batches_tracked[INT64, scalar]
  373. %conv4.0.conv2.1.layers.0.bn.running_mean[FLOAT, 128]
  374. %conv4.0.conv2.1.layers.0.bn.running_var[FLOAT, 128]
  375. %conv4.0.conv2.1.layers.0.bn.weight[FLOAT, 128]
  376. %conv4.0.conv2.1.layers.0.conv1.weight[FLOAT, 128x128x1x1]
  377. %conv4.0.conv2.1.layers.0.conv2.weight[FLOAT, 128x1x3x3]
  378. %conv4.0.conv2.1.layers.1.bn.bias[FLOAT, 128]
  379. %conv4.0.conv2.1.layers.1.bn.num_batches_tracked[INT64, scalar]
  380. %conv4.0.conv2.1.layers.1.bn.running_mean[FLOAT, 128]
  381. %conv4.0.conv2.1.layers.1.bn.running_var[FLOAT, 128]
  382. %conv4.0.conv2.1.layers.1.bn.weight[FLOAT, 128]
  383. %conv4.0.conv2.1.layers.1.conv1.weight[FLOAT, 128x128x1x1]
  384. %conv4.0.conv2.1.layers.1.conv2.weight[FLOAT, 128x1x3x3]
  385. %conv4.0.conv2.2.layers.0.bn.bias[FLOAT, 128]
  386. %conv4.0.conv2.2.layers.0.bn.num_batches_tracked[INT64, scalar]
  387. %conv4.0.conv2.2.layers.0.bn.running_mean[FLOAT, 128]
  388. %conv4.0.conv2.2.layers.0.bn.running_var[FLOAT, 128]
  389. %conv4.0.conv2.2.layers.0.bn.weight[FLOAT, 128]
  390. %conv4.0.conv2.2.layers.0.conv1.weight[FLOAT, 128x128x1x1]
  391. %conv4.0.conv2.2.layers.0.conv2.weight[FLOAT, 128x1x3x3]
  392. %conv4.0.conv2.2.layers.1.bn.bias[FLOAT, 128]
  393. %conv4.0.conv2.2.layers.1.bn.num_batches_tracked[INT64, scalar]
  394. %conv4.0.conv2.2.layers.1.bn.running_mean[FLOAT, 128]
  395. %conv4.0.conv2.2.layers.1.bn.running_var[FLOAT, 128]
  396. %conv4.0.conv2.2.layers.1.bn.weight[FLOAT, 128]
  397. %conv4.0.conv2.2.layers.1.conv1.weight[FLOAT, 128x128x1x1]
  398. %conv4.0.conv2.2.layers.1.conv2.weight[FLOAT, 128x1x3x3]
  399. %conv4.0.conv2.2.layers.2.bn.bias[FLOAT, 128]
  400. %conv4.0.conv2.2.layers.2.bn.num_batches_tracked[INT64, scalar]
  401. %conv4.0.conv2.2.layers.2.bn.running_mean[FLOAT, 128]
  402. %conv4.0.conv2.2.layers.2.bn.running_var[FLOAT, 128]
  403. %conv4.0.conv2.2.layers.2.bn.weight[FLOAT, 128]
  404. %conv4.0.conv2.2.layers.2.conv1.weight[FLOAT, 128x128x1x1]
  405. %conv4.0.conv2.2.layers.2.conv2.weight[FLOAT, 128x1x3x3]
  406. %conv4.0.conv2.3.layers.0.bn.bias[FLOAT, 128]
  407. %conv4.0.conv2.3.layers.0.bn.num_batches_tracked[INT64, scalar]
  408. %conv4.0.conv2.3.layers.0.bn.running_mean[FLOAT, 128]
  409. %conv4.0.conv2.3.layers.0.bn.running_var[FLOAT, 128]
  410. %conv4.0.conv2.3.layers.0.bn.weight[FLOAT, 128]
  411. %conv4.0.conv2.3.layers.0.conv1.weight[FLOAT, 128x128x1x1]
  412. %conv4.0.conv2.3.layers.0.conv2.weight[FLOAT, 128x1x3x3]
  413. %conv4.0.conv2.3.layers.1.bn.bias[FLOAT, 128]
  414. %conv4.0.conv2.3.layers.1.bn.num_batches_tracked[INT64, scalar]
  415. %conv4.0.conv2.3.layers.1.bn.running_mean[FLOAT, 128]
  416. %conv4.0.conv2.3.layers.1.bn.running_var[FLOAT, 128]
  417. %conv4.0.conv2.3.layers.1.bn.weight[FLOAT, 128]
  418. %conv4.0.conv2.3.layers.1.conv1.weight[FLOAT, 128x128x1x1]
  419. %conv4.0.conv2.3.layers.1.conv2.weight[FLOAT, 128x1x3x3]
  420. %conv4.0.conv2.3.layers.2.bn.bias[FLOAT, 128]
  421. %conv4.0.conv2.3.layers.2.bn.num_batches_tracked[INT64, scalar]
  422. %conv4.0.conv2.3.layers.2.bn.running_mean[FLOAT, 128]
  423. %conv4.0.conv2.3.layers.2.bn.running_var[FLOAT, 128]
  424. %conv4.0.conv2.3.layers.2.bn.weight[FLOAT, 128]
  425. %conv4.0.conv2.3.layers.2.conv1.weight[FLOAT, 128x128x1x1]
  426. %conv4.0.conv2.3.layers.2.conv2.weight[FLOAT, 128x1x3x3]
  427. %conv4.0.conv2.3.layers.3.bn.bias[FLOAT, 128]
  428. %conv4.0.conv2.3.layers.3.bn.num_batches_tracked[INT64, scalar]
  429. %conv4.0.conv2.3.layers.3.bn.running_mean[FLOAT, 128]
  430. %conv4.0.conv2.3.layers.3.bn.running_var[FLOAT, 128]
  431. %conv4.0.conv2.3.layers.3.bn.weight[FLOAT, 128]
  432. %conv4.0.conv2.3.layers.3.conv1.weight[FLOAT, 128x128x1x1]
  433. %conv4.0.conv2.3.layers.3.conv2.weight[FLOAT, 128x1x3x3]
  434. %conv4.0.conv3.conv.weight[FLOAT, 512x128x1x1]
  435. %conv4.0.downsample.bn.bias[FLOAT, 512]
  436. %conv4.0.downsample.bn.num_batches_tracked[INT64, scalar]
  437. %conv4.0.downsample.bn.running_mean[FLOAT, 512]
  438. %conv4.0.downsample.bn.running_var[FLOAT, 512]
  439. %conv4.0.downsample.bn.weight[FLOAT, 512]
  440. %conv4.0.downsample.conv.weight[FLOAT, 512x384x1x1]
  441. %conv4.0.gate.fc1.bias[FLOAT, 8]
  442. %conv4.0.gate.fc1.weight[FLOAT, 8x128x1x1]
  443. %conv4.0.gate.fc2.bias[FLOAT, 128]
  444. %conv4.0.gate.fc2.weight[FLOAT, 128x8x1x1]
  445. %conv4.1.conv1.bn.bias[FLOAT, 128]
  446. %conv4.1.conv1.bn.num_batches_tracked[INT64, scalar]
  447. %conv4.1.conv1.bn.running_mean[FLOAT, 128]
  448. %conv4.1.conv1.bn.running_var[FLOAT, 128]
  449. %conv4.1.conv1.bn.weight[FLOAT, 128]
  450. %conv4.1.conv1.conv.weight[FLOAT, 128x512x1x1]
  451. %conv4.1.conv2.0.layers.0.bn.bias[FLOAT, 128]
  452. %conv4.1.conv2.0.layers.0.bn.num_batches_tracked[INT64, scalar]
  453. %conv4.1.conv2.0.layers.0.bn.running_mean[FLOAT, 128]
  454. %conv4.1.conv2.0.layers.0.bn.running_var[FLOAT, 128]
  455. %conv4.1.conv2.0.layers.0.bn.weight[FLOAT, 128]
  456. %conv4.1.conv2.0.layers.0.conv1.weight[FLOAT, 128x128x1x1]
  457. %conv4.1.conv2.0.layers.0.conv2.weight[FLOAT, 128x1x3x3]
  458. %conv4.1.conv2.1.layers.0.bn.bias[FLOAT, 128]
  459. %conv4.1.conv2.1.layers.0.bn.num_batches_tracked[INT64, scalar]
  460. %conv4.1.conv2.1.layers.0.bn.running_mean[FLOAT, 128]
  461. %conv4.1.conv2.1.layers.0.bn.running_var[FLOAT, 128]
  462. %conv4.1.conv2.1.layers.0.bn.weight[FLOAT, 128]
  463. %conv4.1.conv2.1.layers.0.conv1.weight[FLOAT, 128x128x1x1]
  464. %conv4.1.conv2.1.layers.0.conv2.weight[FLOAT, 128x1x3x3]
  465. %conv4.1.conv2.1.layers.1.bn.bias[FLOAT, 128]
  466. %conv4.1.conv2.1.layers.1.bn.num_batches_tracked[INT64, scalar]
  467. %conv4.1.conv2.1.layers.1.bn.running_mean[FLOAT, 128]
  468. %conv4.1.conv2.1.layers.1.bn.running_var[FLOAT, 128]
  469. %conv4.1.conv2.1.layers.1.bn.weight[FLOAT, 128]
  470. %conv4.1.conv2.1.layers.1.conv1.weight[FLOAT, 128x128x1x1]
  471. %conv4.1.conv2.1.layers.1.conv2.weight[FLOAT, 128x1x3x3]
  472. %conv4.1.conv2.2.layers.0.bn.bias[FLOAT, 128]
  473. %conv4.1.conv2.2.layers.0.bn.num_batches_tracked[INT64, scalar]
  474. %conv4.1.conv2.2.layers.0.bn.running_mean[FLOAT, 128]
  475. %conv4.1.conv2.2.layers.0.bn.running_var[FLOAT, 128]
  476. %conv4.1.conv2.2.layers.0.bn.weight[FLOAT, 128]
  477. %conv4.1.conv2.2.layers.0.conv1.weight[FLOAT, 128x128x1x1]
  478. %conv4.1.conv2.2.layers.0.conv2.weight[FLOAT, 128x1x3x3]
  479. %conv4.1.conv2.2.layers.1.bn.bias[FLOAT, 128]
  480. %conv4.1.conv2.2.layers.1.bn.num_batches_tracked[INT64, scalar]
  481. %conv4.1.conv2.2.layers.1.bn.running_mean[FLOAT, 128]
  482. %conv4.1.conv2.2.layers.1.bn.running_var[FLOAT, 128]
  483. %conv4.1.conv2.2.layers.1.bn.weight[FLOAT, 128]
  484. %conv4.1.conv2.2.layers.1.conv1.weight[FLOAT, 128x128x1x1]
  485. %conv4.1.conv2.2.layers.1.conv2.weight[FLOAT, 128x1x3x3]
  486. %conv4.1.conv2.2.layers.2.bn.bias[FLOAT, 128]
  487. %conv4.1.conv2.2.layers.2.bn.num_batches_tracked[INT64, scalar]
  488. %conv4.1.conv2.2.layers.2.bn.running_mean[FLOAT, 128]
  489. %conv4.1.conv2.2.layers.2.bn.running_var[FLOAT, 128]
  490. %conv4.1.conv2.2.layers.2.bn.weight[FLOAT, 128]
  491. %conv4.1.conv2.2.layers.2.conv1.weight[FLOAT, 128x128x1x1]
  492. %conv4.1.conv2.2.layers.2.conv2.weight[FLOAT, 128x1x3x3]
  493. %conv4.1.conv2.3.layers.0.bn.bias[FLOAT, 128]
  494. %conv4.1.conv2.3.layers.0.bn.num_batches_tracked[INT64, scalar]
  495. %conv4.1.conv2.3.layers.0.bn.running_mean[FLOAT, 128]
  496. %conv4.1.conv2.3.layers.0.bn.running_var[FLOAT, 128]
  497. %conv4.1.conv2.3.layers.0.bn.weight[FLOAT, 128]
  498. %conv4.1.conv2.3.layers.0.conv1.weight[FLOAT, 128x128x1x1]
  499. %conv4.1.conv2.3.layers.0.conv2.weight[FLOAT, 128x1x3x3]
  500. %conv4.1.conv2.3.layers.1.bn.bias[FLOAT, 128]
  501. %conv4.1.conv2.3.layers.1.bn.num_batches_tracked[INT64, scalar]
  502. %conv4.1.conv2.3.layers.1.bn.running_mean[FLOAT, 128]
  503. %conv4.1.conv2.3.layers.1.bn.running_var[FLOAT, 128]
  504. %conv4.1.conv2.3.layers.1.bn.weight[FLOAT, 128]
  505. %conv4.1.conv2.3.layers.1.conv1.weight[FLOAT, 128x128x1x1]
  506. %conv4.1.conv2.3.layers.1.conv2.weight[FLOAT, 128x1x3x3]
  507. %conv4.1.conv2.3.layers.2.bn.bias[FLOAT, 128]
  508. %conv4.1.conv2.3.layers.2.bn.num_batches_tracked[INT64, scalar]
  509. %conv4.1.conv2.3.layers.2.bn.running_mean[FLOAT, 128]
  510. %conv4.1.conv2.3.layers.2.bn.running_var[FLOAT, 128]
  511. %conv4.1.conv2.3.layers.2.bn.weight[FLOAT, 128]
  512. %conv4.1.conv2.3.layers.2.conv1.weight[FLOAT, 128x128x1x1]
  513. %conv4.1.conv2.3.layers.2.conv2.weight[FLOAT, 128x1x3x3]
  514. %conv4.1.conv2.3.layers.3.bn.bias[FLOAT, 128]
  515. %conv4.1.conv2.3.layers.3.bn.num_batches_tracked[INT64, scalar]
  516. %conv4.1.conv2.3.layers.3.bn.running_mean[FLOAT, 128]
  517. %conv4.1.conv2.3.layers.3.bn.running_var[FLOAT, 128]
  518. %conv4.1.conv2.3.layers.3.bn.weight[FLOAT, 128]
  519. %conv4.1.conv2.3.layers.3.conv1.weight[FLOAT, 128x128x1x1]
  520. %conv4.1.conv2.3.layers.3.conv2.weight[FLOAT, 128x1x3x3]
  521. %conv4.1.conv3.bn.bias[FLOAT, 512]
  522. %conv4.1.conv3.bn.num_batches_tracked[INT64, scalar]
  523. %conv4.1.conv3.bn.running_mean[FLOAT, 512]
  524. %conv4.1.conv3.bn.running_var[FLOAT, 512]
  525. %conv4.1.conv3.bn.weight[FLOAT, 512]
  526. %conv4.1.conv3.conv.weight[FLOAT, 512x128x1x1]
  527. %conv4.1.gate.fc1.bias[FLOAT, 8]
  528. %conv4.1.gate.fc1.weight[FLOAT, 8x128x1x1]
  529. %conv4.1.gate.fc2.bias[FLOAT, 128]
  530. %conv4.1.gate.fc2.weight[FLOAT, 128x8x1x1]
  531. %conv5.bn.bias[FLOAT, 512]
  532. %conv5.bn.num_batches_tracked[INT64, scalar]
  533. %conv5.bn.running_mean[FLOAT, 512]
  534. %conv5.bn.running_var[FLOAT, 512]
  535. %conv5.bn.weight[FLOAT, 512]
  536. %conv5.conv.weight[FLOAT, 512x512x1x1]
  537. %fc.0.bias[FLOAT, 512]
  538. %fc.0.weight[FLOAT, 512x512]
  539. %fc.1.bias[FLOAT, 512]
  540. %fc.1.num_batches_tracked[INT64, scalar]
  541. %fc.1.running_mean[FLOAT, 512]
  542. %fc.1.running_var[FLOAT, 512]
  543. %fc.1.weight[FLOAT, 512]
  544. %pool2.0.bn.bias[FLOAT, 256]
  545. %pool2.0.bn.num_batches_tracked[INT64, scalar]
  546. %pool2.0.bn.running_mean[FLOAT, 256]
  547. %pool2.0.bn.running_var[FLOAT, 256]
  548. %pool2.0.bn.weight[FLOAT, 256]
  549. %pool2.0.conv.weight[FLOAT, 256x256x1x1]
  550. %pool3.0.bn.bias[FLOAT, 384]
  551. %pool3.0.bn.num_batches_tracked[INT64, scalar]
  552. %pool3.0.bn.running_mean[FLOAT, 384]
  553. %pool3.0.bn.running_var[FLOAT, 384]
  554. %pool3.0.bn.weight[FLOAT, 384]
  555. %pool3.0.conv.weight[FLOAT, 384x384x1x1]
  556. ) {
  557. %553 = Conv[dilations = [1, 1], group = 1, kernel_shape = [7, 7], pads = [3, 3, 3, 3], strides = [2, 2]](%input.1, %conv1.conv.weight)
  558. %554 = InstanceNormalization[epsilon = 9.99999974737875e-06](%553, %conv1.bn.weight, %conv1.bn.bias)
  559. %555 = Relu(%554)
  560. %556 = MaxPool[kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [2, 2]](%555)
  561. %557 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%556, %conv2.0.conv1.conv.weight)
  562. %558 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%557, %conv2.0.conv1.bn.weight, %conv2.0.conv1.bn.bias, %conv2.0.conv1.bn.running_mean, %conv2.0.conv1.bn.running_var)
  563. %559 = Relu(%558)
  564. %560 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%559, %conv2.0.conv2.0.layers.0.conv1.weight)
  565. %561 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%560, %conv2.0.conv2.0.layers.0.conv2.weight)
  566. %562 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%561, %conv2.0.conv2.0.layers.0.bn.weight, %conv2.0.conv2.0.layers.0.bn.bias, %conv2.0.conv2.0.layers.0.bn.running_mean, %conv2.0.conv2.0.layers.0.bn.running_var)
  567. %563 = Relu(%562)
  568. %564 = GlobalAveragePool(%563)
  569. %565 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%564, %conv2.0.gate.fc1.weight, %conv2.0.gate.fc1.bias)
  570. %566 = Relu(%565)
  571. %567 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%566, %conv2.0.gate.fc2.weight, %conv2.0.gate.fc2.bias)
  572. %568 = Sigmoid(%567)
  573. %569 = Mul(%563, %568)
  574. %570 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%559, %conv2.0.conv2.1.layers.0.conv1.weight)
  575. %571 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%570, %conv2.0.conv2.1.layers.0.conv2.weight)
  576. %572 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%571, %conv2.0.conv2.1.layers.0.bn.weight, %conv2.0.conv2.1.layers.0.bn.bias, %conv2.0.conv2.1.layers.0.bn.running_mean, %conv2.0.conv2.1.layers.0.bn.running_var)
  577. %573 = Relu(%572)
  578. %574 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%573, %conv2.0.conv2.1.layers.1.conv1.weight)
  579. %575 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%574, %conv2.0.conv2.1.layers.1.conv2.weight)
  580. %576 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%575, %conv2.0.conv2.1.layers.1.bn.weight, %conv2.0.conv2.1.layers.1.bn.bias, %conv2.0.conv2.1.layers.1.bn.running_mean, %conv2.0.conv2.1.layers.1.bn.running_var)
  581. %577 = Relu(%576)
  582. %578 = GlobalAveragePool(%577)
  583. %579 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%578, %conv2.0.gate.fc1.weight, %conv2.0.gate.fc1.bias)
  584. %580 = Relu(%579)
  585. %581 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%580, %conv2.0.gate.fc2.weight, %conv2.0.gate.fc2.bias)
  586. %582 = Sigmoid(%581)
  587. %583 = Mul(%577, %582)
  588. %584 = Add(%569, %583)
  589. %585 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%559, %conv2.0.conv2.2.layers.0.conv1.weight)
  590. %586 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%585, %conv2.0.conv2.2.layers.0.conv2.weight)
  591. %587 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%586, %conv2.0.conv2.2.layers.0.bn.weight, %conv2.0.conv2.2.layers.0.bn.bias, %conv2.0.conv2.2.layers.0.bn.running_mean, %conv2.0.conv2.2.layers.0.bn.running_var)
  592. %588 = Relu(%587)
  593. %589 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%588, %conv2.0.conv2.2.layers.1.conv1.weight)
  594. %590 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%589, %conv2.0.conv2.2.layers.1.conv2.weight)
  595. %591 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%590, %conv2.0.conv2.2.layers.1.bn.weight, %conv2.0.conv2.2.layers.1.bn.bias, %conv2.0.conv2.2.layers.1.bn.running_mean, %conv2.0.conv2.2.layers.1.bn.running_var)
  596. %592 = Relu(%591)
  597. %593 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%592, %conv2.0.conv2.2.layers.2.conv1.weight)
  598. %594 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%593, %conv2.0.conv2.2.layers.2.conv2.weight)
  599. %595 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%594, %conv2.0.conv2.2.layers.2.bn.weight, %conv2.0.conv2.2.layers.2.bn.bias, %conv2.0.conv2.2.layers.2.bn.running_mean, %conv2.0.conv2.2.layers.2.bn.running_var)
  600. %596 = Relu(%595)
  601. %597 = GlobalAveragePool(%596)
  602. %598 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%597, %conv2.0.gate.fc1.weight, %conv2.0.gate.fc1.bias)
  603. %599 = Relu(%598)
  604. %600 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%599, %conv2.0.gate.fc2.weight, %conv2.0.gate.fc2.bias)
  605. %601 = Sigmoid(%600)
  606. %602 = Mul(%596, %601)
  607. %603 = Add(%584, %602)
  608. %604 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%559, %conv2.0.conv2.3.layers.0.conv1.weight)
  609. %605 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%604, %conv2.0.conv2.3.layers.0.conv2.weight)
  610. %606 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%605, %conv2.0.conv2.3.layers.0.bn.weight, %conv2.0.conv2.3.layers.0.bn.bias, %conv2.0.conv2.3.layers.0.bn.running_mean, %conv2.0.conv2.3.layers.0.bn.running_var)
  611. %607 = Relu(%606)
  612. %608 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%607, %conv2.0.conv2.3.layers.1.conv1.weight)
  613. %609 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%608, %conv2.0.conv2.3.layers.1.conv2.weight)
  614. %610 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%609, %conv2.0.conv2.3.layers.1.bn.weight, %conv2.0.conv2.3.layers.1.bn.bias, %conv2.0.conv2.3.layers.1.bn.running_mean, %conv2.0.conv2.3.layers.1.bn.running_var)
  615. %611 = Relu(%610)
  616. %612 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%611, %conv2.0.conv2.3.layers.2.conv1.weight)
  617. %613 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%612, %conv2.0.conv2.3.layers.2.conv2.weight)
  618. %614 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%613, %conv2.0.conv2.3.layers.2.bn.weight, %conv2.0.conv2.3.layers.2.bn.bias, %conv2.0.conv2.3.layers.2.bn.running_mean, %conv2.0.conv2.3.layers.2.bn.running_var)
  619. %615 = Relu(%614)
  620. %616 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%615, %conv2.0.conv2.3.layers.3.conv1.weight)
  621. %617 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%616, %conv2.0.conv2.3.layers.3.conv2.weight)
  622. %618 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%617, %conv2.0.conv2.3.layers.3.bn.weight, %conv2.0.conv2.3.layers.3.bn.bias, %conv2.0.conv2.3.layers.3.bn.running_mean, %conv2.0.conv2.3.layers.3.bn.running_var)
  623. %619 = Relu(%618)
  624. %620 = GlobalAveragePool(%619)
  625. %621 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%620, %conv2.0.gate.fc1.weight, %conv2.0.gate.fc1.bias)
  626. %622 = Relu(%621)
  627. %623 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%622, %conv2.0.gate.fc2.weight, %conv2.0.gate.fc2.bias)
  628. %624 = Sigmoid(%623)
  629. %625 = Mul(%619, %624)
  630. %626 = Add(%603, %625)
  631. %627 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%626, %conv2.0.conv3.conv.weight)
  632. %628 = InstanceNormalization[epsilon = 9.99999974737875e-06](%627, %conv2.0.IN.weight, %conv2.0.IN.bias)
  633. %629 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%556, %conv2.0.downsample.conv.weight)
  634. %630 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%629, %conv2.0.downsample.bn.weight, %conv2.0.downsample.bn.bias, %conv2.0.downsample.bn.running_mean, %conv2.0.downsample.bn.running_var)
  635. %631 = Add(%628, %630)
  636. %632 = Relu(%631)
  637. %633 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%632, %conv2.1.conv1.conv.weight)
  638. %634 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%633, %conv2.1.conv1.bn.weight, %conv2.1.conv1.bn.bias, %conv2.1.conv1.bn.running_mean, %conv2.1.conv1.bn.running_var)
  639. %635 = Relu(%634)
  640. %636 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%635, %conv2.1.conv2.0.layers.0.conv1.weight)
  641. %637 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%636, %conv2.1.conv2.0.layers.0.conv2.weight)
  642. %638 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%637, %conv2.1.conv2.0.layers.0.bn.weight, %conv2.1.conv2.0.layers.0.bn.bias, %conv2.1.conv2.0.layers.0.bn.running_mean, %conv2.1.conv2.0.layers.0.bn.running_var)
  643. %639 = Relu(%638)
  644. %640 = GlobalAveragePool(%639)
  645. %641 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%640, %conv2.1.gate.fc1.weight, %conv2.1.gate.fc1.bias)
  646. %642 = Relu(%641)
  647. %643 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%642, %conv2.1.gate.fc2.weight, %conv2.1.gate.fc2.bias)
  648. %644 = Sigmoid(%643)
  649. %645 = Mul(%639, %644)
  650. %646 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%635, %conv2.1.conv2.1.layers.0.conv1.weight)
  651. %647 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%646, %conv2.1.conv2.1.layers.0.conv2.weight)
  652. %648 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%647, %conv2.1.conv2.1.layers.0.bn.weight, %conv2.1.conv2.1.layers.0.bn.bias, %conv2.1.conv2.1.layers.0.bn.running_mean, %conv2.1.conv2.1.layers.0.bn.running_var)
  653. %649 = Relu(%648)
  654. %650 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%649, %conv2.1.conv2.1.layers.1.conv1.weight)
  655. %651 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%650, %conv2.1.conv2.1.layers.1.conv2.weight)
  656. %652 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%651, %conv2.1.conv2.1.layers.1.bn.weight, %conv2.1.conv2.1.layers.1.bn.bias, %conv2.1.conv2.1.layers.1.bn.running_mean, %conv2.1.conv2.1.layers.1.bn.running_var)
  657. %653 = Relu(%652)
  658. %654 = GlobalAveragePool(%653)
  659. %655 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%654, %conv2.1.gate.fc1.weight, %conv2.1.gate.fc1.bias)
  660. %656 = Relu(%655)
  661. %657 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%656, %conv2.1.gate.fc2.weight, %conv2.1.gate.fc2.bias)
  662. %658 = Sigmoid(%657)
  663. %659 = Mul(%653, %658)
  664. %660 = Add(%645, %659)
  665. %661 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%635, %conv2.1.conv2.2.layers.0.conv1.weight)
  666. %662 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%661, %conv2.1.conv2.2.layers.0.conv2.weight)
  667. %663 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%662, %conv2.1.conv2.2.layers.0.bn.weight, %conv2.1.conv2.2.layers.0.bn.bias, %conv2.1.conv2.2.layers.0.bn.running_mean, %conv2.1.conv2.2.layers.0.bn.running_var)
  668. %664 = Relu(%663)
  669. %665 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%664, %conv2.1.conv2.2.layers.1.conv1.weight)
  670. %666 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%665, %conv2.1.conv2.2.layers.1.conv2.weight)
  671. %667 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%666, %conv2.1.conv2.2.layers.1.bn.weight, %conv2.1.conv2.2.layers.1.bn.bias, %conv2.1.conv2.2.layers.1.bn.running_mean, %conv2.1.conv2.2.layers.1.bn.running_var)
  672. %668 = Relu(%667)
  673. %669 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%668, %conv2.1.conv2.2.layers.2.conv1.weight)
  674. %670 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%669, %conv2.1.conv2.2.layers.2.conv2.weight)
  675. %671 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%670, %conv2.1.conv2.2.layers.2.bn.weight, %conv2.1.conv2.2.layers.2.bn.bias, %conv2.1.conv2.2.layers.2.bn.running_mean, %conv2.1.conv2.2.layers.2.bn.running_var)
  676. %672 = Relu(%671)
  677. %673 = GlobalAveragePool(%672)
  678. %674 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%673, %conv2.1.gate.fc1.weight, %conv2.1.gate.fc1.bias)
  679. %675 = Relu(%674)
  680. %676 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%675, %conv2.1.gate.fc2.weight, %conv2.1.gate.fc2.bias)
  681. %677 = Sigmoid(%676)
  682. %678 = Mul(%672, %677)
  683. %679 = Add(%660, %678)
  684. %680 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%635, %conv2.1.conv2.3.layers.0.conv1.weight)
  685. %681 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%680, %conv2.1.conv2.3.layers.0.conv2.weight)
  686. %682 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%681, %conv2.1.conv2.3.layers.0.bn.weight, %conv2.1.conv2.3.layers.0.bn.bias, %conv2.1.conv2.3.layers.0.bn.running_mean, %conv2.1.conv2.3.layers.0.bn.running_var)
  687. %683 = Relu(%682)
  688. %684 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%683, %conv2.1.conv2.3.layers.1.conv1.weight)
  689. %685 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%684, %conv2.1.conv2.3.layers.1.conv2.weight)
  690. %686 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%685, %conv2.1.conv2.3.layers.1.bn.weight, %conv2.1.conv2.3.layers.1.bn.bias, %conv2.1.conv2.3.layers.1.bn.running_mean, %conv2.1.conv2.3.layers.1.bn.running_var)
  691. %687 = Relu(%686)
  692. %688 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%687, %conv2.1.conv2.3.layers.2.conv1.weight)
  693. %689 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%688, %conv2.1.conv2.3.layers.2.conv2.weight)
  694. %690 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%689, %conv2.1.conv2.3.layers.2.bn.weight, %conv2.1.conv2.3.layers.2.bn.bias, %conv2.1.conv2.3.layers.2.bn.running_mean, %conv2.1.conv2.3.layers.2.bn.running_var)
  695. %691 = Relu(%690)
  696. %692 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%691, %conv2.1.conv2.3.layers.3.conv1.weight)
  697. %693 = Conv[dilations = [1, 1], group = 64, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%692, %conv2.1.conv2.3.layers.3.conv2.weight)
  698. %694 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%693, %conv2.1.conv2.3.layers.3.bn.weight, %conv2.1.conv2.3.layers.3.bn.bias, %conv2.1.conv2.3.layers.3.bn.running_mean, %conv2.1.conv2.3.layers.3.bn.running_var)
  699. %695 = Relu(%694)
  700. %696 = GlobalAveragePool(%695)
  701. %697 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%696, %conv2.1.gate.fc1.weight, %conv2.1.gate.fc1.bias)
  702. %698 = Relu(%697)
  703. %699 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%698, %conv2.1.gate.fc2.weight, %conv2.1.gate.fc2.bias)
  704. %700 = Sigmoid(%699)
  705. %701 = Mul(%695, %700)
  706. %702 = Add(%679, %701)
  707. %703 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%702, %conv2.1.conv3.conv.weight)
  708. %704 = InstanceNormalization[epsilon = 9.99999974737875e-06](%703, %conv2.1.IN.weight, %conv2.1.IN.bias)
  709. %705 = Add(%704, %632)
  710. %706 = Relu(%705)
  711. %707 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%706, %pool2.0.conv.weight)
  712. %708 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%707, %pool2.0.bn.weight, %pool2.0.bn.bias, %pool2.0.bn.running_mean, %pool2.0.bn.running_var)
  713. %709 = Relu(%708)
  714. %710 = Pad[mode = 'constant', pads = [0, 0, 0, 0, 0, 0, 0, 0], value = 0](%709)
  715. %711 = AveragePool[kernel_shape = [2, 2], pads = [0, 0, 0, 0], strides = [2, 2]](%710)
  716. %712 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%711, %conv3.0.conv1.conv.weight)
  717. %713 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%712, %conv3.0.conv1.bn.weight, %conv3.0.conv1.bn.bias, %conv3.0.conv1.bn.running_mean, %conv3.0.conv1.bn.running_var)
  718. %714 = Relu(%713)
  719. %715 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%714, %conv3.0.conv2.0.layers.0.conv1.weight)
  720. %716 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%715, %conv3.0.conv2.0.layers.0.conv2.weight)
  721. %717 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%716, %conv3.0.conv2.0.layers.0.bn.weight, %conv3.0.conv2.0.layers.0.bn.bias, %conv3.0.conv2.0.layers.0.bn.running_mean, %conv3.0.conv2.0.layers.0.bn.running_var)
  722. %718 = Relu(%717)
  723. %719 = GlobalAveragePool(%718)
  724. %720 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%719, %conv3.0.gate.fc1.weight, %conv3.0.gate.fc1.bias)
  725. %721 = Relu(%720)
  726. %722 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%721, %conv3.0.gate.fc2.weight, %conv3.0.gate.fc2.bias)
  727. %723 = Sigmoid(%722)
  728. %724 = Mul(%718, %723)
  729. %725 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%714, %conv3.0.conv2.1.layers.0.conv1.weight)
  730. %726 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%725, %conv3.0.conv2.1.layers.0.conv2.weight)
  731. %727 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%726, %conv3.0.conv2.1.layers.0.bn.weight, %conv3.0.conv2.1.layers.0.bn.bias, %conv3.0.conv2.1.layers.0.bn.running_mean, %conv3.0.conv2.1.layers.0.bn.running_var)
  732. %728 = Relu(%727)
  733. %729 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%728, %conv3.0.conv2.1.layers.1.conv1.weight)
  734. %730 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%729, %conv3.0.conv2.1.layers.1.conv2.weight)
  735. %731 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%730, %conv3.0.conv2.1.layers.1.bn.weight, %conv3.0.conv2.1.layers.1.bn.bias, %conv3.0.conv2.1.layers.1.bn.running_mean, %conv3.0.conv2.1.layers.1.bn.running_var)
  736. %732 = Relu(%731)
  737. %733 = GlobalAveragePool(%732)
  738. %734 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%733, %conv3.0.gate.fc1.weight, %conv3.0.gate.fc1.bias)
  739. %735 = Relu(%734)
  740. %736 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%735, %conv3.0.gate.fc2.weight, %conv3.0.gate.fc2.bias)
  741. %737 = Sigmoid(%736)
  742. %738 = Mul(%732, %737)
  743. %739 = Add(%724, %738)
  744. %740 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%714, %conv3.0.conv2.2.layers.0.conv1.weight)
  745. %741 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%740, %conv3.0.conv2.2.layers.0.conv2.weight)
  746. %742 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%741, %conv3.0.conv2.2.layers.0.bn.weight, %conv3.0.conv2.2.layers.0.bn.bias, %conv3.0.conv2.2.layers.0.bn.running_mean, %conv3.0.conv2.2.layers.0.bn.running_var)
  747. %743 = Relu(%742)
  748. %744 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%743, %conv3.0.conv2.2.layers.1.conv1.weight)
  749. %745 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%744, %conv3.0.conv2.2.layers.1.conv2.weight)
  750. %746 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%745, %conv3.0.conv2.2.layers.1.bn.weight, %conv3.0.conv2.2.layers.1.bn.bias, %conv3.0.conv2.2.layers.1.bn.running_mean, %conv3.0.conv2.2.layers.1.bn.running_var)
  751. %747 = Relu(%746)
  752. %748 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%747, %conv3.0.conv2.2.layers.2.conv1.weight)
  753. %749 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%748, %conv3.0.conv2.2.layers.2.conv2.weight)
  754. %750 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%749, %conv3.0.conv2.2.layers.2.bn.weight, %conv3.0.conv2.2.layers.2.bn.bias, %conv3.0.conv2.2.layers.2.bn.running_mean, %conv3.0.conv2.2.layers.2.bn.running_var)
  755. %751 = Relu(%750)
  756. %752 = GlobalAveragePool(%751)
  757. %753 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%752, %conv3.0.gate.fc1.weight, %conv3.0.gate.fc1.bias)
  758. %754 = Relu(%753)
  759. %755 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%754, %conv3.0.gate.fc2.weight, %conv3.0.gate.fc2.bias)
  760. %756 = Sigmoid(%755)
  761. %757 = Mul(%751, %756)
  762. %758 = Add(%739, %757)
  763. %759 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%714, %conv3.0.conv2.3.layers.0.conv1.weight)
  764. %760 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%759, %conv3.0.conv2.3.layers.0.conv2.weight)
  765. %761 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%760, %conv3.0.conv2.3.layers.0.bn.weight, %conv3.0.conv2.3.layers.0.bn.bias, %conv3.0.conv2.3.layers.0.bn.running_mean, %conv3.0.conv2.3.layers.0.bn.running_var)
  766. %762 = Relu(%761)
  767. %763 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%762, %conv3.0.conv2.3.layers.1.conv1.weight)
  768. %764 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%763, %conv3.0.conv2.3.layers.1.conv2.weight)
  769. %765 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%764, %conv3.0.conv2.3.layers.1.bn.weight, %conv3.0.conv2.3.layers.1.bn.bias, %conv3.0.conv2.3.layers.1.bn.running_mean, %conv3.0.conv2.3.layers.1.bn.running_var)
  770. %766 = Relu(%765)
  771. %767 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%766, %conv3.0.conv2.3.layers.2.conv1.weight)
  772. %768 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%767, %conv3.0.conv2.3.layers.2.conv2.weight)
  773. %769 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%768, %conv3.0.conv2.3.layers.2.bn.weight, %conv3.0.conv2.3.layers.2.bn.bias, %conv3.0.conv2.3.layers.2.bn.running_mean, %conv3.0.conv2.3.layers.2.bn.running_var)
  774. %770 = Relu(%769)
  775. %771 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%770, %conv3.0.conv2.3.layers.3.conv1.weight)
  776. %772 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%771, %conv3.0.conv2.3.layers.3.conv2.weight)
  777. %773 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%772, %conv3.0.conv2.3.layers.3.bn.weight, %conv3.0.conv2.3.layers.3.bn.bias, %conv3.0.conv2.3.layers.3.bn.running_mean, %conv3.0.conv2.3.layers.3.bn.running_var)
  778. %774 = Relu(%773)
  779. %775 = GlobalAveragePool(%774)
  780. %776 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%775, %conv3.0.gate.fc1.weight, %conv3.0.gate.fc1.bias)
  781. %777 = Relu(%776)
  782. %778 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%777, %conv3.0.gate.fc2.weight, %conv3.0.gate.fc2.bias)
  783. %779 = Sigmoid(%778)
  784. %780 = Mul(%774, %779)
  785. %781 = Add(%758, %780)
  786. %782 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%781, %conv3.0.conv3.conv.weight)
  787. %783 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%782, %conv3.0.conv3.bn.weight, %conv3.0.conv3.bn.bias, %conv3.0.conv3.bn.running_mean, %conv3.0.conv3.bn.running_var)
  788. %784 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%711, %conv3.0.downsample.conv.weight)
  789. %785 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%784, %conv3.0.downsample.bn.weight, %conv3.0.downsample.bn.bias, %conv3.0.downsample.bn.running_mean, %conv3.0.downsample.bn.running_var)
  790. %786 = Add(%783, %785)
  791. %787 = Relu(%786)
  792. %788 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%787, %conv3.1.conv1.conv.weight)
  793. %789 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%788, %conv3.1.conv1.bn.weight, %conv3.1.conv1.bn.bias, %conv3.1.conv1.bn.running_mean, %conv3.1.conv1.bn.running_var)
  794. %790 = Relu(%789)
  795. %791 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%790, %conv3.1.conv2.0.layers.0.conv1.weight)
  796. %792 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%791, %conv3.1.conv2.0.layers.0.conv2.weight)
  797. %793 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%792, %conv3.1.conv2.0.layers.0.bn.weight, %conv3.1.conv2.0.layers.0.bn.bias, %conv3.1.conv2.0.layers.0.bn.running_mean, %conv3.1.conv2.0.layers.0.bn.running_var)
  798. %794 = Relu(%793)
  799. %795 = GlobalAveragePool(%794)
  800. %796 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%795, %conv3.1.gate.fc1.weight, %conv3.1.gate.fc1.bias)
  801. %797 = Relu(%796)
  802. %798 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%797, %conv3.1.gate.fc2.weight, %conv3.1.gate.fc2.bias)
  803. %799 = Sigmoid(%798)
  804. %800 = Mul(%794, %799)
  805. %801 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%790, %conv3.1.conv2.1.layers.0.conv1.weight)
  806. %802 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%801, %conv3.1.conv2.1.layers.0.conv2.weight)
  807. %803 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%802, %conv3.1.conv2.1.layers.0.bn.weight, %conv3.1.conv2.1.layers.0.bn.bias, %conv3.1.conv2.1.layers.0.bn.running_mean, %conv3.1.conv2.1.layers.0.bn.running_var)
  808. %804 = Relu(%803)
  809. %805 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%804, %conv3.1.conv2.1.layers.1.conv1.weight)
  810. %806 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%805, %conv3.1.conv2.1.layers.1.conv2.weight)
  811. %807 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%806, %conv3.1.conv2.1.layers.1.bn.weight, %conv3.1.conv2.1.layers.1.bn.bias, %conv3.1.conv2.1.layers.1.bn.running_mean, %conv3.1.conv2.1.layers.1.bn.running_var)
  812. %808 = Relu(%807)
  813. %809 = GlobalAveragePool(%808)
  814. %810 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%809, %conv3.1.gate.fc1.weight, %conv3.1.gate.fc1.bias)
  815. %811 = Relu(%810)
  816. %812 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%811, %conv3.1.gate.fc2.weight, %conv3.1.gate.fc2.bias)
  817. %813 = Sigmoid(%812)
  818. %814 = Mul(%808, %813)
  819. %815 = Add(%800, %814)
  820. %816 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%790, %conv3.1.conv2.2.layers.0.conv1.weight)
  821. %817 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%816, %conv3.1.conv2.2.layers.0.conv2.weight)
  822. %818 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%817, %conv3.1.conv2.2.layers.0.bn.weight, %conv3.1.conv2.2.layers.0.bn.bias, %conv3.1.conv2.2.layers.0.bn.running_mean, %conv3.1.conv2.2.layers.0.bn.running_var)
  823. %819 = Relu(%818)
  824. %820 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%819, %conv3.1.conv2.2.layers.1.conv1.weight)
  825. %821 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%820, %conv3.1.conv2.2.layers.1.conv2.weight)
  826. %822 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%821, %conv3.1.conv2.2.layers.1.bn.weight, %conv3.1.conv2.2.layers.1.bn.bias, %conv3.1.conv2.2.layers.1.bn.running_mean, %conv3.1.conv2.2.layers.1.bn.running_var)
  827. %823 = Relu(%822)
  828. %824 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%823, %conv3.1.conv2.2.layers.2.conv1.weight)
  829. %825 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%824, %conv3.1.conv2.2.layers.2.conv2.weight)
  830. %826 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%825, %conv3.1.conv2.2.layers.2.bn.weight, %conv3.1.conv2.2.layers.2.bn.bias, %conv3.1.conv2.2.layers.2.bn.running_mean, %conv3.1.conv2.2.layers.2.bn.running_var)
  831. %827 = Relu(%826)
  832. %828 = GlobalAveragePool(%827)
  833. %829 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%828, %conv3.1.gate.fc1.weight, %conv3.1.gate.fc1.bias)
  834. %830 = Relu(%829)
  835. %831 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%830, %conv3.1.gate.fc2.weight, %conv3.1.gate.fc2.bias)
  836. %832 = Sigmoid(%831)
  837. %833 = Mul(%827, %832)
  838. %834 = Add(%815, %833)
  839. %835 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%790, %conv3.1.conv2.3.layers.0.conv1.weight)
  840. %836 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%835, %conv3.1.conv2.3.layers.0.conv2.weight)
  841. %837 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%836, %conv3.1.conv2.3.layers.0.bn.weight, %conv3.1.conv2.3.layers.0.bn.bias, %conv3.1.conv2.3.layers.0.bn.running_mean, %conv3.1.conv2.3.layers.0.bn.running_var)
  842. %838 = Relu(%837)
  843. %839 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%838, %conv3.1.conv2.3.layers.1.conv1.weight)
  844. %840 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%839, %conv3.1.conv2.3.layers.1.conv2.weight)
  845. %841 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%840, %conv3.1.conv2.3.layers.1.bn.weight, %conv3.1.conv2.3.layers.1.bn.bias, %conv3.1.conv2.3.layers.1.bn.running_mean, %conv3.1.conv2.3.layers.1.bn.running_var)
  846. %842 = Relu(%841)
  847. %843 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%842, %conv3.1.conv2.3.layers.2.conv1.weight)
  848. %844 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%843, %conv3.1.conv2.3.layers.2.conv2.weight)
  849. %845 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%844, %conv3.1.conv2.3.layers.2.bn.weight, %conv3.1.conv2.3.layers.2.bn.bias, %conv3.1.conv2.3.layers.2.bn.running_mean, %conv3.1.conv2.3.layers.2.bn.running_var)
  850. %846 = Relu(%845)
  851. %847 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%846, %conv3.1.conv2.3.layers.3.conv1.weight)
  852. %848 = Conv[dilations = [1, 1], group = 96, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%847, %conv3.1.conv2.3.layers.3.conv2.weight)
  853. %849 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%848, %conv3.1.conv2.3.layers.3.bn.weight, %conv3.1.conv2.3.layers.3.bn.bias, %conv3.1.conv2.3.layers.3.bn.running_mean, %conv3.1.conv2.3.layers.3.bn.running_var)
  854. %850 = Relu(%849)
  855. %851 = GlobalAveragePool(%850)
  856. %852 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%851, %conv3.1.gate.fc1.weight, %conv3.1.gate.fc1.bias)
  857. %853 = Relu(%852)
  858. %854 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%853, %conv3.1.gate.fc2.weight, %conv3.1.gate.fc2.bias)
  859. %855 = Sigmoid(%854)
  860. %856 = Mul(%850, %855)
  861. %857 = Add(%834, %856)
  862. %858 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%857, %conv3.1.conv3.conv.weight)
  863. %859 = InstanceNormalization[epsilon = 9.99999974737875e-06](%858, %conv3.1.IN.weight, %conv3.1.IN.bias)
  864. %860 = Add(%859, %787)
  865. %861 = Relu(%860)
  866. %862 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%861, %pool3.0.conv.weight)
  867. %863 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%862, %pool3.0.bn.weight, %pool3.0.bn.bias, %pool3.0.bn.running_mean, %pool3.0.bn.running_var)
  868. %864 = Relu(%863)
  869. %865 = Pad[mode = 'constant', pads = [0, 0, 0, 0, 0, 0, 0, 0], value = 0](%864)
  870. %866 = AveragePool[kernel_shape = [2, 2], pads = [0, 0, 0, 0], strides = [2, 2]](%865)
  871. %867 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%866, %conv4.0.conv1.conv.weight)
  872. %868 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%867, %conv4.0.conv1.bn.weight, %conv4.0.conv1.bn.bias, %conv4.0.conv1.bn.running_mean, %conv4.0.conv1.bn.running_var)
  873. %869 = Relu(%868)
  874. %870 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%869, %conv4.0.conv2.0.layers.0.conv1.weight)
  875. %871 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%870, %conv4.0.conv2.0.layers.0.conv2.weight)
  876. %872 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%871, %conv4.0.conv2.0.layers.0.bn.weight, %conv4.0.conv2.0.layers.0.bn.bias, %conv4.0.conv2.0.layers.0.bn.running_mean, %conv4.0.conv2.0.layers.0.bn.running_var)
  877. %873 = Relu(%872)
  878. %874 = GlobalAveragePool(%873)
  879. %875 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%874, %conv4.0.gate.fc1.weight, %conv4.0.gate.fc1.bias)
  880. %876 = Relu(%875)
  881. %877 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%876, %conv4.0.gate.fc2.weight, %conv4.0.gate.fc2.bias)
  882. %878 = Sigmoid(%877)
  883. %879 = Mul(%873, %878)
  884. %880 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%869, %conv4.0.conv2.1.layers.0.conv1.weight)
  885. %881 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%880, %conv4.0.conv2.1.layers.0.conv2.weight)
  886. %882 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%881, %conv4.0.conv2.1.layers.0.bn.weight, %conv4.0.conv2.1.layers.0.bn.bias, %conv4.0.conv2.1.layers.0.bn.running_mean, %conv4.0.conv2.1.layers.0.bn.running_var)
  887. %883 = Relu(%882)
  888. %884 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%883, %conv4.0.conv2.1.layers.1.conv1.weight)
  889. %885 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%884, %conv4.0.conv2.1.layers.1.conv2.weight)
  890. %886 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%885, %conv4.0.conv2.1.layers.1.bn.weight, %conv4.0.conv2.1.layers.1.bn.bias, %conv4.0.conv2.1.layers.1.bn.running_mean, %conv4.0.conv2.1.layers.1.bn.running_var)
  891. %887 = Relu(%886)
  892. %888 = GlobalAveragePool(%887)
  893. %889 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%888, %conv4.0.gate.fc1.weight, %conv4.0.gate.fc1.bias)
  894. %890 = Relu(%889)
  895. %891 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%890, %conv4.0.gate.fc2.weight, %conv4.0.gate.fc2.bias)
  896. %892 = Sigmoid(%891)
  897. %893 = Mul(%887, %892)
  898. %894 = Add(%879, %893)
  899. %895 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%869, %conv4.0.conv2.2.layers.0.conv1.weight)
  900. %896 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%895, %conv4.0.conv2.2.layers.0.conv2.weight)
  901. %897 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%896, %conv4.0.conv2.2.layers.0.bn.weight, %conv4.0.conv2.2.layers.0.bn.bias, %conv4.0.conv2.2.layers.0.bn.running_mean, %conv4.0.conv2.2.layers.0.bn.running_var)
  902. %898 = Relu(%897)
  903. %899 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%898, %conv4.0.conv2.2.layers.1.conv1.weight)
  904. %900 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%899, %conv4.0.conv2.2.layers.1.conv2.weight)
  905. %901 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%900, %conv4.0.conv2.2.layers.1.bn.weight, %conv4.0.conv2.2.layers.1.bn.bias, %conv4.0.conv2.2.layers.1.bn.running_mean, %conv4.0.conv2.2.layers.1.bn.running_var)
  906. %902 = Relu(%901)
  907. %903 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%902, %conv4.0.conv2.2.layers.2.conv1.weight)
  908. %904 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%903, %conv4.0.conv2.2.layers.2.conv2.weight)
  909. %905 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%904, %conv4.0.conv2.2.layers.2.bn.weight, %conv4.0.conv2.2.layers.2.bn.bias, %conv4.0.conv2.2.layers.2.bn.running_mean, %conv4.0.conv2.2.layers.2.bn.running_var)
  910. %906 = Relu(%905)
  911. %907 = GlobalAveragePool(%906)
  912. %908 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%907, %conv4.0.gate.fc1.weight, %conv4.0.gate.fc1.bias)
  913. %909 = Relu(%908)
  914. %910 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%909, %conv4.0.gate.fc2.weight, %conv4.0.gate.fc2.bias)
  915. %911 = Sigmoid(%910)
  916. %912 = Mul(%906, %911)
  917. %913 = Add(%894, %912)
  918. %914 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%869, %conv4.0.conv2.3.layers.0.conv1.weight)
  919. %915 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%914, %conv4.0.conv2.3.layers.0.conv2.weight)
  920. %916 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%915, %conv4.0.conv2.3.layers.0.bn.weight, %conv4.0.conv2.3.layers.0.bn.bias, %conv4.0.conv2.3.layers.0.bn.running_mean, %conv4.0.conv2.3.layers.0.bn.running_var)
  921. %917 = Relu(%916)
  922. %918 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%917, %conv4.0.conv2.3.layers.1.conv1.weight)
  923. %919 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%918, %conv4.0.conv2.3.layers.1.conv2.weight)
  924. %920 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%919, %conv4.0.conv2.3.layers.1.bn.weight, %conv4.0.conv2.3.layers.1.bn.bias, %conv4.0.conv2.3.layers.1.bn.running_mean, %conv4.0.conv2.3.layers.1.bn.running_var)
  925. %921 = Relu(%920)
  926. %922 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%921, %conv4.0.conv2.3.layers.2.conv1.weight)
  927. %923 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%922, %conv4.0.conv2.3.layers.2.conv2.weight)
  928. %924 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%923, %conv4.0.conv2.3.layers.2.bn.weight, %conv4.0.conv2.3.layers.2.bn.bias, %conv4.0.conv2.3.layers.2.bn.running_mean, %conv4.0.conv2.3.layers.2.bn.running_var)
  929. %925 = Relu(%924)
  930. %926 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%925, %conv4.0.conv2.3.layers.3.conv1.weight)
  931. %927 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%926, %conv4.0.conv2.3.layers.3.conv2.weight)
  932. %928 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%927, %conv4.0.conv2.3.layers.3.bn.weight, %conv4.0.conv2.3.layers.3.bn.bias, %conv4.0.conv2.3.layers.3.bn.running_mean, %conv4.0.conv2.3.layers.3.bn.running_var)
  933. %929 = Relu(%928)
  934. %930 = GlobalAveragePool(%929)
  935. %931 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%930, %conv4.0.gate.fc1.weight, %conv4.0.gate.fc1.bias)
  936. %932 = Relu(%931)
  937. %933 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%932, %conv4.0.gate.fc2.weight, %conv4.0.gate.fc2.bias)
  938. %934 = Sigmoid(%933)
  939. %935 = Mul(%929, %934)
  940. %936 = Add(%913, %935)
  941. %937 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%936, %conv4.0.conv3.conv.weight)
  942. %938 = InstanceNormalization[epsilon = 9.99999974737875e-06](%937, %conv4.0.IN.weight, %conv4.0.IN.bias)
  943. %939 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%866, %conv4.0.downsample.conv.weight)
  944. %940 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%939, %conv4.0.downsample.bn.weight, %conv4.0.downsample.bn.bias, %conv4.0.downsample.bn.running_mean, %conv4.0.downsample.bn.running_var)
  945. %941 = Add(%938, %940)
  946. %942 = Relu(%941)
  947. %943 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%942, %conv4.1.conv1.conv.weight)
  948. %944 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%943, %conv4.1.conv1.bn.weight, %conv4.1.conv1.bn.bias, %conv4.1.conv1.bn.running_mean, %conv4.1.conv1.bn.running_var)
  949. %945 = Relu(%944)
  950. %946 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%945, %conv4.1.conv2.0.layers.0.conv1.weight)
  951. %947 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%946, %conv4.1.conv2.0.layers.0.conv2.weight)
  952. %948 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%947, %conv4.1.conv2.0.layers.0.bn.weight, %conv4.1.conv2.0.layers.0.bn.bias, %conv4.1.conv2.0.layers.0.bn.running_mean, %conv4.1.conv2.0.layers.0.bn.running_var)
  953. %949 = Relu(%948)
  954. %950 = GlobalAveragePool(%949)
  955. %951 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%950, %conv4.1.gate.fc1.weight, %conv4.1.gate.fc1.bias)
  956. %952 = Relu(%951)
  957. %953 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%952, %conv4.1.gate.fc2.weight, %conv4.1.gate.fc2.bias)
  958. %954 = Sigmoid(%953)
  959. %955 = Mul(%949, %954)
  960. %956 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%945, %conv4.1.conv2.1.layers.0.conv1.weight)
  961. %957 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%956, %conv4.1.conv2.1.layers.0.conv2.weight)
  962. %958 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%957, %conv4.1.conv2.1.layers.0.bn.weight, %conv4.1.conv2.1.layers.0.bn.bias, %conv4.1.conv2.1.layers.0.bn.running_mean, %conv4.1.conv2.1.layers.0.bn.running_var)
  963. %959 = Relu(%958)
  964. %960 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%959, %conv4.1.conv2.1.layers.1.conv1.weight)
  965. %961 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%960, %conv4.1.conv2.1.layers.1.conv2.weight)
  966. %962 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%961, %conv4.1.conv2.1.layers.1.bn.weight, %conv4.1.conv2.1.layers.1.bn.bias, %conv4.1.conv2.1.layers.1.bn.running_mean, %conv4.1.conv2.1.layers.1.bn.running_var)
  967. %963 = Relu(%962)
  968. %964 = GlobalAveragePool(%963)
  969. %965 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%964, %conv4.1.gate.fc1.weight, %conv4.1.gate.fc1.bias)
  970. %966 = Relu(%965)
  971. %967 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%966, %conv4.1.gate.fc2.weight, %conv4.1.gate.fc2.bias)
  972. %968 = Sigmoid(%967)
  973. %969 = Mul(%963, %968)
  974. %970 = Add(%955, %969)
  975. %971 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%945, %conv4.1.conv2.2.layers.0.conv1.weight)
  976. %972 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%971, %conv4.1.conv2.2.layers.0.conv2.weight)
  977. %973 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%972, %conv4.1.conv2.2.layers.0.bn.weight, %conv4.1.conv2.2.layers.0.bn.bias, %conv4.1.conv2.2.layers.0.bn.running_mean, %conv4.1.conv2.2.layers.0.bn.running_var)
  978. %974 = Relu(%973)
  979. %975 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%974, %conv4.1.conv2.2.layers.1.conv1.weight)
  980. %976 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%975, %conv4.1.conv2.2.layers.1.conv2.weight)
  981. %977 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%976, %conv4.1.conv2.2.layers.1.bn.weight, %conv4.1.conv2.2.layers.1.bn.bias, %conv4.1.conv2.2.layers.1.bn.running_mean, %conv4.1.conv2.2.layers.1.bn.running_var)
  982. %978 = Relu(%977)
  983. %979 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%978, %conv4.1.conv2.2.layers.2.conv1.weight)
  984. %980 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%979, %conv4.1.conv2.2.layers.2.conv2.weight)
  985. %981 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%980, %conv4.1.conv2.2.layers.2.bn.weight, %conv4.1.conv2.2.layers.2.bn.bias, %conv4.1.conv2.2.layers.2.bn.running_mean, %conv4.1.conv2.2.layers.2.bn.running_var)
  986. %982 = Relu(%981)
  987. %983 = GlobalAveragePool(%982)
  988. %984 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%983, %conv4.1.gate.fc1.weight, %conv4.1.gate.fc1.bias)
  989. %985 = Relu(%984)
  990. %986 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%985, %conv4.1.gate.fc2.weight, %conv4.1.gate.fc2.bias)
  991. %987 = Sigmoid(%986)
  992. %988 = Mul(%982, %987)
  993. %989 = Add(%970, %988)
  994. %990 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%945, %conv4.1.conv2.3.layers.0.conv1.weight)
  995. %991 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%990, %conv4.1.conv2.3.layers.0.conv2.weight)
  996. %992 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%991, %conv4.1.conv2.3.layers.0.bn.weight, %conv4.1.conv2.3.layers.0.bn.bias, %conv4.1.conv2.3.layers.0.bn.running_mean, %conv4.1.conv2.3.layers.0.bn.running_var)
  997. %993 = Relu(%992)
  998. %994 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%993, %conv4.1.conv2.3.layers.1.conv1.weight)
  999. %995 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%994, %conv4.1.conv2.3.layers.1.conv2.weight)
  1000. %996 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%995, %conv4.1.conv2.3.layers.1.bn.weight, %conv4.1.conv2.3.layers.1.bn.bias, %conv4.1.conv2.3.layers.1.bn.running_mean, %conv4.1.conv2.3.layers.1.bn.running_var)
  1001. %997 = Relu(%996)
  1002. %998 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%997, %conv4.1.conv2.3.layers.2.conv1.weight)
  1003. %999 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%998, %conv4.1.conv2.3.layers.2.conv2.weight)
  1004. %1000 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%999, %conv4.1.conv2.3.layers.2.bn.weight, %conv4.1.conv2.3.layers.2.bn.bias, %conv4.1.conv2.3.layers.2.bn.running_mean, %conv4.1.conv2.3.layers.2.bn.running_var)
  1005. %1001 = Relu(%1000)
  1006. %1002 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%1001, %conv4.1.conv2.3.layers.3.conv1.weight)
  1007. %1003 = Conv[dilations = [1, 1], group = 128, kernel_shape = [3, 3], pads = [1, 1, 1, 1], strides = [1, 1]](%1002, %conv4.1.conv2.3.layers.3.conv2.weight)
  1008. %1004 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%1003, %conv4.1.conv2.3.layers.3.bn.weight, %conv4.1.conv2.3.layers.3.bn.bias, %conv4.1.conv2.3.layers.3.bn.running_mean, %conv4.1.conv2.3.layers.3.bn.running_var)
  1009. %1005 = Relu(%1004)
  1010. %1006 = GlobalAveragePool(%1005)
  1011. %1007 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%1006, %conv4.1.gate.fc1.weight, %conv4.1.gate.fc1.bias)
  1012. %1008 = Relu(%1007)
  1013. %1009 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%1008, %conv4.1.gate.fc2.weight, %conv4.1.gate.fc2.bias)
  1014. %1010 = Sigmoid(%1009)
  1015. %1011 = Mul(%1005, %1010)
  1016. %1012 = Add(%989, %1011)
  1017. %1013 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%1012, %conv4.1.conv3.conv.weight)
  1018. %1014 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%1013, %conv4.1.conv3.bn.weight, %conv4.1.conv3.bn.bias, %conv4.1.conv3.bn.running_mean, %conv4.1.conv3.bn.running_var)
  1019. %1015 = Add(%1014, %942)
  1020. %1016 = Relu(%1015)
  1021. %1017 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%1016, %conv5.conv.weight)
  1022. %1018 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%1017, %conv5.bn.weight, %conv5.bn.bias, %conv5.bn.running_mean, %conv5.bn.running_var)
  1023. %1019 = Relu(%1018)
  1024. %1020 = GlobalAveragePool(%1019)
  1025. %1021 = Shape(%1020)
  1026. %1022 = Constant[value = <Scalar Tensor []>]()
  1027. %1023 = Gather[axis = 0](%1021, %1022)
  1028. %1024 = Constant[value = <Scalar Tensor []>]()
  1029. %1025 = Unsqueeze[axes = [0]](%1023)
  1030. %1026 = Unsqueeze[axes = [0]](%1024)
  1031. %1027 = Concat[axis = 0](%1025, %1026)
  1032. %1028 = Reshape(%1020, %1027)
  1033. %1029 = Gemm[alpha = 1, beta = 1, transB = 1](%1028, %fc.0.weight, %fc.0.bias)
  1034. %1030 = BatchNormalization[epsilon = 9.99999974737875e-06, momentum = 0.899999976158142](%1029, %fc.1.weight, %fc.1.bias, %fc.1.running_mean, %fc.1.running_var)
  1035. %1031 = Relu(%1030)
  1036. return %1031
  1037. }
Advertisement
Add Comment
Please, Sign In to add comment
Public Pastes
We use cookies for various purposes including analytics. By continuing to use Pastebin, you agree to our use of cookies as described in the Cookies Policy.  OK, I Understand
Not a member of Pastebin yet?
Sign Up, it unlocks many cool features!