Toybrick

标题: rv1126模型推理数据异常 [打印本页]

作者: tree    时间: 2023-2-9 19:28
标题: rv1126模型推理数据异常
版本说明:
    rv1126驱动版本1.7.0
    rknn-toolkit版本1.7.0
    pytorch 版本1.9.0 和1.6.0
    onnx 版本 1.6.0

流程:
    使用pytorch训练模型-->转成onnx 格式-->转换为指定平台的rknn模型-->模型转移到板子上拿到推理结果-->计算onnx推理结果与板子上模型的推理结果的余弦相似度。

问题:
    1.官网的demo onnx 模型转换后,推理结果相似度很高,0.9以上,例如yolov5,但是我训练出来的模型转换后相似度不足0.5,而且对比了模型的一个卷积层的输出,发现相似度也只有0.5左右,最后将模型转为了rk1808平台的模型,推理结果与onnx模型的相似度又达到了0.9以上。对比rknn-toolkit1.7.0 的文档,将所有的三方包版本都对应上了,依旧没能解决这个问题。


最后感谢大佬相助!


作者: tree    时间: 2023-2-9 19:30
这个问题困扰了我一周了,尝试了各种方案,多种模型,多个版本的pytorch和onnx依旧没有解决
作者: tree    时间: 2023-2-9 20:19
附上一份转换yolov3-tiny的log
  1. D Save log info to: log/best_rv1126_sample_asymmetric_affine-u8_True_2023-02-09_02:43:42.log
  2. D Using CPPUTILS: True
  3. I Start importing onnx...
  4. W Call onnx.optimizer.optimize fail, skip optimize
  5. I Current ONNX Model use ir_version 6 opset_version 11
  6. I Call RKNN onnx optimize fail, skip optimize
  7. D Calc tensor Constant_66 (0,)
  8. D Calc tensor Initializer_model.8.conv.weight (256, 128, 3, 3)
  9. D Calc tensor Initializer_model.8.conv.bias (256,)
  10. D Calc tensor Initializer_model.6.conv.weight (128, 64, 3, 3)
  11. D Calc tensor Initializer_model.6.conv.bias (128,)
  12. D Calc tensor Initializer_model.4.conv.weight (64, 32, 3, 3)
  13. D Calc tensor Initializer_model.4.conv.bias (64,)
  14. D Calc tensor Initializer_model.2.conv.weight (32, 16, 3, 3)
  15. D Calc tensor Initializer_model.2.conv.bias (32,)
  16. D Calc tensor Initializer_model.19.conv.weight (256, 384, 3, 3)
  17. D Calc tensor Initializer_model.19.conv.bias (256,)
  18. D Calc tensor Initializer_model.16.conv.weight (128, 256, 1, 1)
  19. D Calc tensor Initializer_model.16.conv.bias (128,)
  20. D Calc tensor Initializer_model.15.conv.weight (512, 256, 3, 3)
  21. D Calc tensor Initializer_model.15.conv.bias (512,)
  22. D Calc tensor Initializer_model.14.conv.weight (256, 1024, 1, 1)
  23. D Calc tensor Initializer_model.14.conv.bias (256,)
  24. D Calc tensor Initializer_model.13.conv.weight (1024, 512, 3, 3)
  25. D Calc tensor Initializer_model.13.conv.bias (1024,)
  26. D Calc tensor Initializer_model.11.conv.weight (512, 512, 2, 2)
  27. D Calc tensor Initializer_model.11.conv.bias (512,)
  28. D Calc tensor Initializer_model.10.conv.weight (512, 256, 3, 3)
  29. D Calc tensor Initializer_model.10.conv.bias (512,)
  30. D Calc tensor Initializer_model.0.conv.weight (16, 3, 3, 3)
  31. D Calc tensor Initializer_model.0.conv.bias (16,)
  32. D Calc tensor Conv_30 (1, 16, 640, 640)
  33. D Calc tensor Relu_31 (1, 16, 640, 640)
  34. D Calc tensor MaxPool_32 (1, 16, 320, 320)
  35. D Calc tensor Conv_33 (1, 32, 320, 320)
  36. D Calc tensor Relu_34 (1, 32, 320, 320)
  37. D Calc tensor MaxPool_35 (1, 32, 160, 160)
  38. D Calc tensor Conv_36 (1, 64, 160, 160)
  39. D Calc tensor Relu_37 (1, 64, 160, 160)
  40. D Calc tensor MaxPool_38 (1, 64, 80, 80)
  41. D Calc tensor Conv_39 (1, 128, 80, 80)
  42. D Calc tensor Relu_40 (1, 128, 80, 80)
  43. D Calc tensor MaxPool_41 (1, 128, 40, 40)
  44. D Calc tensor Conv_42 (1, 256, 40, 40)
  45. D Calc tensor Relu_43 (1, 256, 40, 40)
  46. D Calc tensor MaxPool_44 (1, 256, 20, 20)
  47. D Calc tensor Conv_45 (1, 512, 20, 20)
  48. D Calc tensor Relu_46 (1, 512, 20, 20)
  49. D Calc tensor Conv_47 (1, 512, 21, 21)
  50. D Calc tensor Relu_48 (1, 512, 21, 21)
  51. D Calc tensor MaxPool_49 (1, 512, 20, 20)
  52. D Calc tensor Conv_50 (1, 1024, 20, 20)
  53. D Calc tensor Relu_51 (1, 1024, 20, 20)
  54. D Calc tensor Conv_52 (1, 256, 20, 20)
  55. D Calc tensor Relu_53 (1, 256, 20, 20)
  56. D Calc tensor Conv_54 (1, 512, 20, 20)
  57. D Calc tensor Relu_55 (1, 512, 20, 20)
  58. D Calc tensor Conv_56 (1, 128, 20, 20)
  59. D Calc tensor Relu_57 (1, 128, 20, 20)
  60. D Calc tensor Initializer_75 (4,)
  61. D Calc tensor Resize_67 (1, 128, 40, 40)
  62. D Calc tensor Concat_68 (1, 384, 40, 40)
  63. D Calc tensor Conv_69 (1, 256, 40, 40)
  64. D Calc tensor Relu_output (1, 256, 40, 40)
  65. D import clients finished
  66. I build output layer attach_Relu_Relu_32:out0
  67. I build output layer attach_Relu_Relu_25:out0
  68. I Try match Relu_Relu_25:out0
  69. I Match r_relu [['Relu_Relu_25']] [['Relu']] to [['relu']]
  70. I Try match Relu_Relu_32:out0
  71. I Match r_relu [['Relu_Relu_32']] [['Relu']] to [['relu']]
  72. I Try match Conv_Conv_24:out0
  73. I Match r_conv [['Conv_Conv_24', 'Initializer_model.15.conv.weight', 'Initializer_model.15.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  74. I Try match Conv_Conv_31:out0
  75. I Match r_conv [['Conv_Conv_31', 'Initializer_model.19.conv.weight', 'Initializer_model.19.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  76. I Try match Relu_Relu_23:out0
  77. I Match r_relu [['Relu_Relu_23']] [['Relu']] to [['relu']]
  78. I Try match Concat_Concat_30:out0
  79. I Match concat_2 [['Concat_Concat_30']] [['Concat']] to [['concat']]
  80. I Try match Conv_Conv_22:out0
  81. I Match r_conv [['Conv_Conv_22', 'Initializer_model.14.conv.weight', 'Initializer_model.14.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  82. I Try match Resize_Resize_29:out0
  83. I Match r_resize [['Resize_Resize_29', 'Constant_Constant_28', 'Initializer_75']] [['Resize', 'Constant', 'Constant_1']] to [['image_resize']]
  84. I Try match Relu_Relu_13:out0
  85. I Match r_relu [['Relu_Relu_13']] [['Relu']] to [['relu']]
  86. I Try match Relu_Relu_21:out0
  87. I Match r_relu [['Relu_Relu_21']] [['Relu']] to [['relu']]
  88. I Try match Relu_Relu_27:out0
  89. I Match r_relu [['Relu_Relu_27']] [['Relu']] to [['relu']]
  90. I Try match Conv_Conv_12:out0
  91. I Match r_conv [['Conv_Conv_12', 'Initializer_model.8.conv.weight', 'Initializer_model.8.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  92. I Try match Conv_Conv_20:out0
  93. I Match r_conv [['Conv_Conv_20', 'Initializer_model.13.conv.weight', 'Initializer_model.13.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  94. I Try match Conv_Conv_26:out0
  95. I Match r_conv [['Conv_Conv_26', 'Initializer_model.16.conv.weight', 'Initializer_model.16.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  96. I Try match MaxPool_MaxPool_11:out0
  97. I Match r_maxpool [['MaxPool_MaxPool_11']] [['MaxPool']] to [['pooling']]
  98. I Try match MaxPool_MaxPool_19:out0
  99. I Match r_maxpool [['MaxPool_MaxPool_19']] [['MaxPool']] to [['pooling']]
  100. I Try match Relu_Relu_18:out0
  101. I Match r_relu [['Relu_Relu_18']] [['Relu']] to [['relu']]
  102. I Try match Relu_Relu_10:out0
  103. I Match r_relu [['Relu_Relu_10']] [['Relu']] to [['relu']]
  104. I Try match Conv_Conv_17:out0
  105. I Match r_conv [['Conv_Conv_17', 'Initializer_model.11.conv.weight', 'Initializer_model.11.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  106. I Try match Conv_Conv_9:out0
  107. I Match r_conv [['Conv_Conv_9', 'Initializer_model.6.conv.weight', 'Initializer_model.6.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  108. I Try match Relu_Relu_16:out0
  109. I Match r_relu [['Relu_Relu_16']] [['Relu']] to [['relu']]
  110. I Try match MaxPool_MaxPool_8:out0
  111. I Match r_maxpool [['MaxPool_MaxPool_8']] [['MaxPool']] to [['pooling']]
  112. I Try match Conv_Conv_15:out0
  113. I Match r_conv [['Conv_Conv_15', 'Initializer_model.10.conv.weight', 'Initializer_model.10.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  114. I Try match Relu_Relu_7:out0
  115. I Match r_relu [['Relu_Relu_7']] [['Relu']] to [['relu']]
  116. I Try match MaxPool_MaxPool_14:out0
  117. I Match r_maxpool [['MaxPool_MaxPool_14']] [['MaxPool']] to [['pooling']]
  118. I Try match Conv_Conv_6:out0
  119. I Match r_conv [['Conv_Conv_6', 'Initializer_model.4.conv.weight', 'Initializer_model.4.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  120. I Try match MaxPool_MaxPool_5:out0
  121. I Match r_maxpool [['MaxPool_MaxPool_5']] [['MaxPool']] to [['pooling']]
  122. I Try match Relu_Relu_4:out0
  123. I Match r_relu [['Relu_Relu_4']] [['Relu']] to [['relu']]
  124. I Try match Conv_Conv_3:out0
  125. I Match r_conv [['Conv_Conv_3', 'Initializer_model.2.conv.weight', 'Initializer_model.2.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  126. I Try match MaxPool_MaxPool_2:out0
  127. I Match r_maxpool [['MaxPool_MaxPool_2']] [['MaxPool']] to [['pooling']]
  128. I Try match Relu_Relu_1:out0
  129. I Match r_relu [['Relu_Relu_1']] [['Relu']] to [['relu']]
  130. I Try match Conv_Conv_0:out0
  131. I Match r_conv [['Conv_Conv_0', 'Initializer_model.0.conv.weight', 'Initializer_model.0.conv.bias']] [['Conv', 'Constant_0', 'Constant_1']] to [['convolution']]
  132. I build input layer images:out0
  133. D connect Conv_Conv_24_4 0  ~ Relu_Relu_25_2 0
  134. D connect Conv_Conv_31_5 0  ~ Relu_Relu_32_3 0
  135. D connect Relu_Relu_23_6 0  ~ Conv_Conv_24_4 0
  136. D connect Concat_Concat_30_7 0  ~ Conv_Conv_31_5 0
  137. D connect Conv_Conv_22_8 0  ~ Relu_Relu_23_6 0
  138. D connect Resize_Resize_29_9 0  ~ Concat_Concat_30_7 0
  139. D connect Relu_Relu_13_10 0  ~ Concat_Concat_30_7 1
  140. D connect Relu_Relu_21_11 0  ~ Conv_Conv_22_8 0
  141. D connect Relu_Relu_27_12 0  ~ Resize_Resize_29_9 0
  142. D connect Conv_Conv_12_13 0  ~ Relu_Relu_13_10 0
  143. D connect Conv_Conv_20_14 0  ~ Relu_Relu_21_11 0
  144. D connect Conv_Conv_26_15 0  ~ Relu_Relu_27_12 0
  145. D connect MaxPool_MaxPool_11_16 0  ~ Conv_Conv_12_13 0
  146. D connect MaxPool_MaxPool_19_17 0  ~ Conv_Conv_20_14 0
  147. D connect Relu_Relu_23_6 0  ~ Conv_Conv_26_15 0
  148. D connect Relu_Relu_10_19 0  ~ MaxPool_MaxPool_11_16 0
  149. D connect Relu_Relu_18_18 0  ~ MaxPool_MaxPool_19_17 0
  150. D connect Conv_Conv_17_20 0  ~ Relu_Relu_18_18 0
  151. D connect Conv_Conv_9_21 0  ~ Relu_Relu_10_19 0
  152. D connect Relu_Relu_16_22 0  ~ Conv_Conv_17_20 0
  153. D connect MaxPool_MaxPool_8_23 0  ~ Conv_Conv_9_21 0
  154. D connect Conv_Conv_15_24 0  ~ Relu_Relu_16_22 0
  155. D connect Relu_Relu_7_25 0  ~ MaxPool_MaxPool_8_23 0
  156. D connect MaxPool_MaxPool_14_26 0  ~ Conv_Conv_15_24 0
  157. D connect Conv_Conv_6_27 0  ~ Relu_Relu_7_25 0
  158. D connect Relu_Relu_13_10 0  ~ MaxPool_MaxPool_14_26 0
  159. D connect MaxPool_MaxPool_5_28 0  ~ Conv_Conv_6_27 0
  160. D connect Relu_Relu_4_29 0  ~ MaxPool_MaxPool_5_28 0
  161. D connect Conv_Conv_3_30 0  ~ Relu_Relu_4_29 0
  162. D connect MaxPool_MaxPool_2_31 0  ~ Conv_Conv_3_30 0
  163. D connect Relu_Relu_1_32 0  ~ MaxPool_MaxPool_2_31 0
  164. D connect Conv_Conv_0_33 0  ~ Relu_Relu_1_32 0
  165. D connect images_34 0  ~ Conv_Conv_0_33 0
  166. D connect Relu_Relu_32_3 0  ~ attach_Relu_Relu_32/out0_0 0
  167. D connect Relu_Relu_25_2 0  ~ attach_Relu_Relu_25/out0_1 0
  168. D Process images_34 ...
  169. D RKNN output shape(input): (1 640 640 3)
  170. D Tensor @images_34:out0 type: float32
  171. D Process Conv_Conv_0_33 ...
  172. D RKNN output shape(convolution): (1 640 640 16)
  173. D Tensor @Conv_Conv_0_33:out0 type: float32
  174. D Process Relu_Relu_1_32 ...
  175. D RKNN output shape(relu): (1 640 640 16)
  176. D Tensor @Relu_Relu_1_32:out0 type: float32
  177. D Process MaxPool_MaxPool_2_31 ...
  178. D RKNN output shape(pooling): (1 320 320 16)
  179. D Tensor @MaxPool_MaxPool_2_31:out0 type: float32
  180. D Process Conv_Conv_3_30 ...
  181. D RKNN output shape(convolution): (1 320 320 32)
  182. D Tensor @Conv_Conv_3_30:out0 type: float32
  183. D Process Relu_Relu_4_29 ...
  184. D RKNN output shape(relu): (1 320 320 32)
  185. D Tensor @Relu_Relu_4_29:out0 type: float32
  186. D Process MaxPool_MaxPool_5_28 ...
  187. D RKNN output shape(pooling): (1 160 160 32)
  188. D Tensor @MaxPool_MaxPool_5_28:out0 type: float32
  189. D Process Conv_Conv_6_27 ...
  190. D RKNN output shape(convolution): (1 160 160 64)
  191. D Tensor @Conv_Conv_6_27:out0 type: float32
  192. D Process Relu_Relu_7_25 ...
  193. D RKNN output shape(relu): (1 160 160 64)
  194. D Tensor @Relu_Relu_7_25:out0 type: float32
  195. D Process MaxPool_MaxPool_8_23 ...
  196. D RKNN output shape(pooling): (1 80 80 64)
  197. D Tensor @MaxPool_MaxPool_8_23:out0 type: float32
  198. D Process Conv_Conv_9_21 ...
  199. D RKNN output shape(convolution): (1 80 80 128)
  200. D Tensor @Conv_Conv_9_21:out0 type: float32
  201. D Process Relu_Relu_10_19 ...
  202. D RKNN output shape(relu): (1 80 80 128)
  203. D Tensor @Relu_Relu_10_19:out0 type: float32
  204. D Process MaxPool_MaxPool_11_16 ...
  205. D RKNN output shape(pooling): (1 40 40 128)
  206. D Tensor @MaxPool_MaxPool_11_16:out0 type: float32
  207. D Process Conv_Conv_12_13 ...
  208. D RKNN output shape(convolution): (1 40 40 256)
  209. D Tensor @Conv_Conv_12_13:out0 type: float32
  210. D Process Relu_Relu_13_10 ...
  211. D RKNN output shape(relu): (1 40 40 256)
  212. D Tensor @Relu_Relu_13_10:out0 type: float32
  213. D Process MaxPool_MaxPool_14_26 ...
  214. D RKNN output shape(pooling): (1 20 20 256)
  215. D Tensor @MaxPool_MaxPool_14_26:out0 type: float32
  216. D Process Conv_Conv_15_24 ...
  217. D RKNN output shape(convolution): (1 20 20 512)
  218. D Tensor @Conv_Conv_15_24:out0 type: float32
  219. D Process Relu_Relu_16_22 ...
  220. D RKNN output shape(relu): (1 20 20 512)
  221. D Tensor @Relu_Relu_16_22:out0 type: float32
  222. D Process Conv_Conv_17_20 ...
  223. D RKNN output shape(convolution): (1 21 21 512)
  224. D Tensor @Conv_Conv_17_20:out0 type: float32
  225. D Process Relu_Relu_18_18 ...
  226. D RKNN output shape(relu): (1 21 21 512)
  227. D Tensor @Relu_Relu_18_18:out0 type: float32
  228. D Process MaxPool_MaxPool_19_17 ...
  229. D RKNN output shape(pooling): (1 20 20 512)
  230. D Tensor @MaxPool_MaxPool_19_17:out0 type: float32
  231. D Process Conv_Conv_20_14 ...
  232. D RKNN output shape(convolution): (1 20 20 1024)
  233. D Tensor @Conv_Conv_20_14:out0 type: float32
  234. D Process Relu_Relu_21_11 ...
  235. D RKNN output shape(relu): (1 20 20 1024)
  236. D Tensor @Relu_Relu_21_11:out0 type: float32
  237. D Process Conv_Conv_22_8 ...
  238. D RKNN output shape(convolution): (1 20 20 256)
  239. D Tensor @Conv_Conv_22_8:out0 type: float32
  240. D Process Relu_Relu_23_6 ...
  241. D RKNN output shape(relu): (1 20 20 256)
  242. D Tensor @Relu_Relu_23_6:out0 type: float32
  243. D Process Conv_Conv_26_15 ...
  244. D RKNN output shape(convolution): (1 20 20 128)
  245. D Tensor @Conv_Conv_26_15:out0 type: float32
  246. D Process Relu_Relu_27_12 ...
  247. D RKNN output shape(relu): (1 20 20 128)
  248. D Tensor @Relu_Relu_27_12:out0 type: float32
  249. D Process Resize_Resize_29_9 ...
  250. D RKNN output shape(image_resize): (1 40 40 128)
  251. D Tensor @Resize_Resize_29_9:out0 type: float32
  252. D Process Concat_Concat_30_7 ...
  253. D RKNN output shape(concat): (1 40 40 384)
  254. D Tensor @Concat_Concat_30_7:out0 type: float32
  255. D Process Conv_Conv_31_5 ...
  256. D RKNN output shape(convolution): (1 40 40 256)
  257. D Tensor @Conv_Conv_31_5:out0 type: float32
  258. D Process Relu_Relu_32_3 ...
  259. D RKNN output shape(relu): (1 40 40 256)
  260. D Tensor @Relu_Relu_32_3:out0 type: float32
  261. D Process attach_Relu_Relu_32/out0_0 ...
  262. D RKNN output shape(output): (1 40 40 256)
  263. D Tensor @attach_Relu_Relu_32/out0_0:out0 type: float32
  264. D Process Conv_Conv_24_4 ...
  265. D RKNN output shape(convolution): (1 20 20 512)
  266. D Tensor @Conv_Conv_24_4:out0 type: float32
  267. D Process Relu_Relu_25_2 ...
  268. D RKNN output shape(relu): (1 20 20 512)
  269. D Tensor @Relu_Relu_25_2:out0 type: float32
  270. D Process attach_Relu_Relu_25/out0_1 ...
  271. D RKNN output shape(output): (1 20 20 512)
  272. D Tensor @attach_Relu_Relu_25/out0_1:out0 type: float32
  273. I Build torch-jit-export complete.
  274. I Start C2T Switcher...
  275. D Optimizing network with broadcast_op
  276. D convert Concat_Concat_30_7(concat) axis 1 to 3
  277. I End C2T Switcher...
  278. D Optimizing network with force_1d_tensor, swapper, merge_duplicate_quantize_dequantize, merge_layer, auto_fill_bn, auto_fill_l2normalizescale, auto_fill_instancenormalize, resize_nearest_transformer, eltwise_transform, auto_fill_multiply, auto_fill_zero_bias, proposal_opt_import, special_add_to_conv2d
  279. I End importing onnx...
  280. W The RKNN Model generated can not run on simulator when pre_compile is True.
  281. I Generate input meta ...
  282. I Load input meta
  283. I Generate input meta ...
  284. D import clients finished
  285. I Load net...
  286. I Load data...
  287. I Load input meta
  288. I Start quantization...
  289. D import clients finished
  290. D iterations: 7, batch_size: 16
  291. I Quantization start...
  292. D set up a quantize net
  293. D *********** Setup input meta ***********
  294. D import clients finished
  295. D *********** Setup database (1) ***********
  296. D Setup provider layer "text_input_layer":
  297. D Lids: ['images_34']
  298. D Shapes: [[16, 640, 640, 3]]
  299. D Data types: ['float32']
  300. D Sparse tensors: []
  301. D Tensor names(H5FS only): []
  302. D Add preprocess "[('reverse_channel', False), ('mean', [0, 0, 0]), ('scale', [0.00392156862745098, 0.00392156862745098, 0.00392156862745098]), ('preproc_node_params', ordereddict([('add_preproc_node', False), ('preproc_type', 'IMAGE_RGB'), ('preproc_perm', [0, 1, 2, 3])]))]" for "images_34"
  303. D *********** Setup input meta complete ***********
  304. D Process images_34 ...
  305. D RKNN output shape(input): (16 640 640 3)
  306. D Tensor @images_34:out0 type: asymmetric_affine
  307. D Real output shape: (16, 640, 640, 3)
  308. D Process Conv_Conv_0_33 ...
  309. D RKNN output shape(convolution): (16 640 640 16)
  310. D Tensor @Conv_Conv_0_33:out0 type: asymmetric_affine
  311. D Real output shape: (16, 640, 640, 16)
  312. D Process Relu_Relu_1_32 ...
  313. D RKNN output shape(relu): (16 640 640 16)
  314. D Tensor @Relu_Relu_1_32:out0 type: asymmetric_affine
  315. D Real output shape: (16, 640, 640, 16)
  316. D Process MaxPool_MaxPool_2_31 ...
  317. D RKNN output shape(pooling): (16 320 320 16)
  318. D Tensor @MaxPool_MaxPool_2_31:out0 type: asymmetric_affine
  319. D Real output shape: (16, 320, 320, 16)
  320. D Process Conv_Conv_3_30 ...
  321. D RKNN output shape(convolution): (16 320 320 32)
  322. D Tensor @Conv_Conv_3_30:out0 type: asymmetric_affine
  323. D Real output shape: (16, 320, 320, 32)
  324. D Process Relu_Relu_4_29 ...
  325. D RKNN output shape(relu): (16 320 320 32)
  326. D Tensor @Relu_Relu_4_29:out0 type: asymmetric_affine
  327. D Real output shape: (16, 320, 320, 32)
  328. D Process MaxPool_MaxPool_5_28 ...
  329. D RKNN output shape(pooling): (16 160 160 32)
  330. D Tensor @MaxPool_MaxPool_5_28:out0 type: asymmetric_affine
  331. D Real output shape: (16, 160, 160, 32)
  332. D Process Conv_Conv_6_27 ...
  333. D RKNN output shape(convolution): (16 160 160 64)
  334. D Tensor @Conv_Conv_6_27:out0 type: asymmetric_affine
  335. D Real output shape: (16, 160, 160, 64)
  336. D Process Relu_Relu_7_25 ...
  337. D RKNN output shape(relu): (16 160 160 64)
  338. D Tensor @Relu_Relu_7_25:out0 type: asymmetric_affine
  339. D Real output shape: (16, 160, 160, 64)
  340. D Process MaxPool_MaxPool_8_23 ...
  341. D RKNN output shape(pooling): (16 80 80 64)
  342. D Tensor @MaxPool_MaxPool_8_23:out0 type: asymmetric_affine
  343. D Real output shape: (16, 80, 80, 64)
  344. D Process Conv_Conv_9_21 ...
  345. D RKNN output shape(convolution): (16 80 80 128)
  346. D Tensor @Conv_Conv_9_21:out0 type: asymmetric_affine
  347. D Real output shape: (16, 80, 80, 128)
  348. D Process Relu_Relu_10_19 ...
  349. D RKNN output shape(relu): (16 80 80 128)
  350. D Tensor @Relu_Relu_10_19:out0 type: asymmetric_affine
  351. D Real output shape: (16, 80, 80, 128)
  352. D Process MaxPool_MaxPool_11_16 ...
  353. D RKNN output shape(pooling): (16 40 40 128)
  354. D Tensor @MaxPool_MaxPool_11_16:out0 type: asymmetric_affine
  355. D Real output shape: (16, 40, 40, 128)
  356. D Process Conv_Conv_12_13 ...
  357. D RKNN output shape(convolution): (16 40 40 256)
  358. D Tensor @Conv_Conv_12_13:out0 type: asymmetric_affine
  359. D Real output shape: (16, 40, 40, 256)
  360. D Process Relu_Relu_13_10 ...
  361. D RKNN output shape(relu): (16 40 40 256)
  362. D Tensor @Relu_Relu_13_10:out0 type: asymmetric_affine
  363. D Real output shape: (16, 40, 40, 256)
  364. D Process MaxPool_MaxPool_14_26 ...
  365. D RKNN output shape(pooling): (16 20 20 256)
  366. D Tensor @MaxPool_MaxPool_14_26:out0 type: asymmetric_affine
  367. D Real output shape: (16, 20, 20, 256)
  368. D Process Conv_Conv_15_24 ...
  369. D RKNN output shape(convolution): (16 20 20 512)
  370. D Tensor @Conv_Conv_15_24:out0 type: asymmetric_affine
  371. D Real output shape: (16, 20, 20, 512)
  372. D Process Relu_Relu_16_22 ...
  373. D RKNN output shape(relu): (16 20 20 512)
  374. D Tensor @Relu_Relu_16_22:out0 type: asymmetric_affine
  375. D Real output shape: (16, 20, 20, 512)
  376. D Process Conv_Conv_17_20 ...
  377. D RKNN output shape(convolution): (16 21 21 512)
  378. D Tensor @Conv_Conv_17_20:out0 type: asymmetric_affine
  379. D Real output shape: (16, 21, 21, 512)
  380. D Process Relu_Relu_18_18 ...
  381. D RKNN output shape(relu): (16 21 21 512)
  382. D Tensor @Relu_Relu_18_18:out0 type: asymmetric_affine
  383. D Real output shape: (16, 21, 21, 512)
  384. D Process MaxPool_MaxPool_19_17 ...
  385. D RKNN output shape(pooling): (16 20 20 512)
  386. D Tensor @MaxPool_MaxPool_19_17:out0 type: asymmetric_affine
  387. D Real output shape: (16, 20, 20, 512)
  388. D Process Conv_Conv_20_14 ...
  389. D RKNN output shape(convolution): (16 20 20 1024)
  390. D Tensor @Conv_Conv_20_14:out0 type: asymmetric_affine
  391. D Real output shape: (16, 20, 20, 1024)
  392. D Process Relu_Relu_21_11 ...
  393. D RKNN output shape(relu): (16 20 20 1024)
  394. D Tensor @Relu_Relu_21_11:out0 type: asymmetric_affine
  395. D Real output shape: (16, 20, 20, 1024)
  396. D Process Conv_Conv_22_8 ...
  397. D RKNN output shape(convolution): (16 20 20 256)
  398. D Tensor @Conv_Conv_22_8:out0 type: asymmetric_affine
  399. D Real output shape: (16, 20, 20, 256)
  400. D Process Relu_Relu_23_6 ...
  401. D RKNN output shape(relu): (16 20 20 256)
  402. D Tensor @Relu_Relu_23_6:out0 type: asymmetric_affine
  403. D Real output shape: (16, 20, 20, 256)
  404. D Process Conv_Conv_26_15 ...
  405. D RKNN output shape(convolution): (16 20 20 128)
  406. D Tensor @Conv_Conv_26_15:out0 type: asymmetric_affine
  407. D Real output shape: (16, 20, 20, 128)
  408. D Process Relu_Relu_27_12 ...
  409. D RKNN output shape(relu): (16 20 20 128)
  410. D Tensor @Relu_Relu_27_12:out0 type: asymmetric_affine
  411. D Real output shape: (16, 20, 20, 128)
  412. D Process Resize_Resize_29_9 ...
  413. D RKNN output shape(image_resize): (16 40 40 128)
  414. D Tensor @Resize_Resize_29_9:out0 type: float32
  415. D Real output shape: (16, 40, 40, 128)
  416. D Process Concat_Concat_30_7 ...
  417. D RKNN output shape(concat): (16 40 40 384)
  418. D Tensor @Concat_Concat_30_7:out0 type: asymmetric_affine
  419. D Real output shape: (16, 40, 40, 384)
  420. D Process Conv_Conv_31_5 ...
  421. D RKNN output shape(convolution): (16 40 40 256)
  422. D Tensor @Conv_Conv_31_5:out0 type: asymmetric_affine
  423. D Real output shape: (16, 40, 40, 256)
  424. D Process Relu_Relu_32_3 ...
  425. D RKNN output shape(relu): (16 40 40 256)
  426. D Tensor @Relu_Relu_32_3:out0 type: asymmetric_affine
  427. D Real output shape: (16, 40, 40, 256)
  428. D Process attach_Relu_Relu_32/out0_0 ...
  429. D RKNN output shape(output): (16 40 40 256)
  430. D Tensor @attach_Relu_Relu_32/out0_0:out0 type: float32
  431. D Real output shape: (16, 40, 40, 256)
  432. D Process Conv_Conv_24_4 ...
  433. D RKNN output shape(convolution): (16 20 20 512)
  434. D Tensor @Conv_Conv_24_4:out0 type: asymmetric_affine
  435. D Real output shape: (16, 20, 20, 512)
  436. D Process Relu_Relu_25_2 ...
  437. D RKNN output shape(relu): (16 20 20 512)
  438. D Tensor @Relu_Relu_25_2:out0 type: asymmetric_affine
  439. D Real output shape: (16, 20, 20, 512)
  440. D Process attach_Relu_Relu_25/out0_1 ...
  441. D RKNN output shape(output): (16 20 20 512)
  442. D Tensor @attach_Relu_Relu_25/out0_1:out0 type: float32
  443. D Real output shape: (16, 20, 20, 512)
  444. I Build torch-jit-export complete.
  445. I Analyze activation range of layer: Conv_Conv_0_33
  446. I Analyze activation range of layer: Relu_Relu_1_32
  447. I Analyze activation range of layer: MaxPool_MaxPool_2_31
  448. I Analyze activation range of layer: Conv_Conv_3_30
  449. I Analyze activation range of layer: Relu_Relu_4_29
  450. I Analyze activation range of layer: MaxPool_MaxPool_5_28
  451. I Analyze activation range of layer: Conv_Conv_6_27
  452. I Analyze activation range of layer: Relu_Relu_7_25
  453. I Analyze activation range of layer: MaxPool_MaxPool_8_23
  454. I Analyze activation range of layer: Conv_Conv_9_21
  455. I Analyze activation range of layer: Relu_Relu_10_19
  456. I Analyze activation range of layer: MaxPool_MaxPool_11_16
  457. I Analyze activation range of layer: Conv_Conv_12_13
  458. I Analyze activation range of layer: Relu_Relu_13_10
  459. I Analyze activation range of layer: MaxPool_MaxPool_14_26
  460. I Analyze activation range of layer: Conv_Conv_15_24
  461. I Analyze activation range of layer: Relu_Relu_16_22
  462. I Analyze activation range of layer: Conv_Conv_17_20
  463. I Analyze activation range of layer: Relu_Relu_18_18
  464. I Analyze activation range of layer: MaxPool_MaxPool_19_17
  465. I Analyze activation range of layer: Conv_Conv_20_14
  466. I Analyze activation range of layer: Relu_Relu_21_11
  467. I Analyze activation range of layer: Conv_Conv_22_8
  468. I Analyze activation range of layer: Relu_Relu_23_6
  469. I Analyze activation range of layer: Conv_Conv_24_4
  470. I Analyze activation range of layer: Conv_Conv_26_15
  471. I Analyze activation range of layer: Relu_Relu_25_2
  472. I Analyze activation range of layer: Relu_Relu_27_12
  473. I Analyze activation range of layer: attach_Relu_Relu_25/out0_1
  474. I Analyze activation range of layer: Resize_Resize_29_9
  475. I Analyze activation range of layer: Concat_Concat_30_7
  476. I Analyze activation range of layer: Conv_Conv_31_5
  477. I Analyze activation range of layer: Relu_Relu_32_3
  478. I Analyze activation range of layer: attach_Relu_Relu_32/out0_0
  479. I Running 7 iterations
  480. D 0(14.28%), Queue size 16
  481. D 1(28.57%), Queue size 16
  482. D 2(42.85%), Queue size 16
  483. D 3(57.14%), Queue size 16
  484. D 4(71.42%), Queue size 16
  485. D 5(85.71%), Queue size 16
  486. D 6(100.00%), Queue size 16
  487. I Queue cancelled.
  488. D Quantize tensor @Relu_Relu_25_2:out0.
  489. D Quantize tensor @Relu_Relu_32_3:out0.
  490. D Quantize tensor @Conv_Conv_24_4:out0.
  491. D Quantize tensor @Conv_Conv_31_5:out0.
  492. D Quantize tensor @Relu_Relu_23_6:out0.
  493. D Quantize tensor @Concat_Concat_30_7:out0.
  494. D Quantize tensor @Conv_Conv_22_8:out0.
  495. D Quantize tensor @Relu_Relu_13_10:out0.
  496. D Quantize tensor @Relu_Relu_21_11:out0.
  497. D Quantize tensor @Relu_Relu_27_12:out0.
  498. D Quantize tensor @Conv_Conv_12_13:out0.
  499. D Quantize tensor @Conv_Conv_20_14:out0.
  500. D Quantize tensor @Conv_Conv_26_15:out0.
  501. D Quantize tensor @MaxPool_MaxPool_11_16:out0.
  502. D Quantize tensor @MaxPool_MaxPool_19_17:out0.
  503. D Quantize tensor @Relu_Relu_18_18:out0.
  504. D Quantize tensor @Relu_Relu_10_19:out0.
  505. D Quantize tensor @Conv_Conv_17_20:out0.
  506. D Quantize tensor @Conv_Conv_9_21:out0.
  507. D Quantize tensor @Relu_Relu_16_22:out0.
  508. D Quantize tensor @MaxPool_MaxPool_8_23:out0.
  509. D Quantize tensor @Conv_Conv_15_24:out0.
  510. D Quantize tensor @Relu_Relu_7_25:out0.
  511. D Quantize tensor @MaxPool_MaxPool_14_26:out0.
  512. D Quantize tensor @Conv_Conv_6_27:out0.
  513. D Quantize tensor @MaxPool_MaxPool_5_28:out0.
  514. D Quantize tensor @Relu_Relu_4_29:out0.
  515. D Quantize tensor @Conv_Conv_3_30:out0.
  516. D Quantize tensor @MaxPool_MaxPool_2_31:out0.
  517. D Quantize tensor @Relu_Relu_1_32:out0.
  518. D Quantize tensor @Conv_Conv_0_33:out0.
  519. D Quantize tensor @images_34:out0.
  520. D Quantize tensor @Conv_Conv_24_4:weight.
  521. D Quantize tensor @Conv_Conv_31_5:weight.
  522. D Quantize tensor @Conv_Conv_22_8:weight.
  523. D Quantize tensor @Conv_Conv_12_13:weight.
  524. D Quantize tensor @Conv_Conv_20_14:weight.
  525. D Quantize tensor @Conv_Conv_26_15:weight.
  526. D Quantize tensor @Conv_Conv_17_20:weight.
  527. D Quantize tensor @Conv_Conv_9_21:weight.
  528. D Quantize tensor @Conv_Conv_15_24:weight.
  529. D Quantize tensor @Conv_Conv_6_27:weight.
  530. D Quantize tensor @Conv_Conv_3_30:weight.
  531. D Quantize tensor @Conv_Conv_0_33:weight.
  532. D Quantize tensor @Conv_Conv_24_4:bias.
  533. D Quantize tensor @Conv_Conv_31_5:bias.
  534. D Quantize tensor @Conv_Conv_22_8:bias.
  535. D Quantize tensor @Conv_Conv_12_13:bias.
  536. D Quantize tensor @Conv_Conv_20_14:bias.
  537. D Quantize tensor @Conv_Conv_26_15:bias.
  538. D Quantize tensor @Conv_Conv_17_20:bias.
  539. D Quantize tensor @Conv_Conv_9_21:bias.
  540. D Quantize tensor @Conv_Conv_15_24:bias.
  541. D Quantize tensor @Conv_Conv_6_27:bias.
  542. D Quantize tensor @Conv_Conv_3_30:bias.
  543. D Quantize tensor @Conv_Conv_0_33:bias.
  544. I Clean.
  545. D Optimizing network with align_quantize, broadcast_quantize, qnt_adjust_coef, qnt_adjust_param
  546. D Align @Relu_Relu_27_12:out0 scale to [0.05264517]
  547. D Align @Concat_Concat_30_7:out0 scale to [0.05264517]
  548. D Align @Relu_Relu_13_10:out0 scale to [0.05264517]
  549. D Quantize tensor(@attach_Relu_Relu_25/out0_1:out0) with tensor(@Relu_Relu_25_2:out0)
  550. D Quantize tensor(@Resize_Resize_29_9:out0) with tensor(@Relu_Relu_27_12:out0)
  551. D Quantize tensor(@attach_Relu_Relu_32/out0_0:out0) with tensor(@Relu_Relu_32_3:out0)
  552. I Conv_Conv_0_33 merged relu nexted.
  553. I Conv_Conv_3_30 merged relu nexted.
  554. I Conv_Conv_6_27 merged relu nexted.
  555. I Conv_Conv_9_21 merged relu nexted.
  556. I Conv_Conv_15_24 merged relu nexted.
  557. I Conv_Conv_17_20 merged relu nexted.
  558. I Conv_Conv_20_14 merged relu nexted.
  559. I Conv_Conv_22_8 merged relu nexted.
  560. I Conv_Conv_24_4 merged relu nexted.
  561. I Conv_Conv_26_15 merged relu nexted.
  562. I Conv_Conv_31_5 merged relu nexted.
  563. I Quantization complete.
  564. I End quantization...
  565. D import clients finished
  566. I Load net...
  567. I Load data...
  568. I Load quantization tensor table
  569. I Load input meta
  570. D Process images_34 ...
  571. D RKNN output shape(input): (1 640 640 3)
  572. D Tensor @images_34:out0 type: asymmetric_affine
  573. D Process Conv_Conv_0_33 ...
  574. D RKNN output shape(convolution): (1 640 640 16)
  575. D Tensor @Conv_Conv_0_33:out0 type: asymmetric_affine
  576. D Process Relu_Relu_1_32 ...
  577. D RKNN output shape(relu): (1 640 640 16)
  578. D Tensor @Relu_Relu_1_32:out0 type: asymmetric_affine
  579. D Process MaxPool_MaxPool_2_31 ...
  580. D RKNN output shape(pooling): (1 320 320 16)
  581. D Tensor @MaxPool_MaxPool_2_31:out0 type: asymmetric_affine
  582. D Process Conv_Conv_3_30 ...
  583. D RKNN output shape(convolution): (1 320 320 32)
  584. D Tensor @Conv_Conv_3_30:out0 type: asymmetric_affine
  585. D Process Relu_Relu_4_29 ...
  586. D RKNN output shape(relu): (1 320 320 32)
  587. D Tensor @Relu_Relu_4_29:out0 type: asymmetric_affine
  588. D Process MaxPool_MaxPool_5_28 ...
  589. D RKNN output shape(pooling): (1 160 160 32)
  590. D Tensor @MaxPool_MaxPool_5_28:out0 type: asymmetric_affine
  591. D Process Conv_Conv_6_27 ...
  592. D RKNN output shape(convolution): (1 160 160 64)
  593. D Tensor @Conv_Conv_6_27:out0 type: asymmetric_affine
  594. D Process Relu_Relu_7_25 ...
  595. D RKNN output shape(relu): (1 160 160 64)
  596. D Tensor @Relu_Relu_7_25:out0 type: asymmetric_affine
  597. D Process MaxPool_MaxPool_8_23 ...
  598. D RKNN output shape(pooling): (1 80 80 64)
  599. D Tensor @MaxPool_MaxPool_8_23:out0 type: asymmetric_affine
  600. D Process Conv_Conv_9_21 ...
  601. D RKNN output shape(convolution): (1 80 80 128)
  602. D Tensor @Conv_Conv_9_21:out0 type: asymmetric_affine
  603. D Process Relu_Relu_10_19 ...
  604. D RKNN output shape(relu): (1 80 80 128)
  605. D Tensor @Relu_Relu_10_19:out0 type: asymmetric_affine
  606. D Process MaxPool_MaxPool_11_16 ...
  607. D RKNN output shape(pooling): (1 40 40 128)
  608. D Tensor @MaxPool_MaxPool_11_16:out0 type: asymmetric_affine
  609. D Process Conv_Conv_12_13 ...
  610. D RKNN output shape(convolution): (1 40 40 256)
  611. D Tensor @Conv_Conv_12_13:out0 type: asymmetric_affine
  612. D Process Relu_Relu_13_10 ...
  613. D RKNN output shape(relu): (1 40 40 256)
  614. D Tensor @Relu_Relu_13_10:out0 type: asymmetric_affine
  615. D Process MaxPool_MaxPool_14_26 ...
  616. D RKNN output shape(pooling): (1 20 20 256)
  617. D Tensor @MaxPool_MaxPool_14_26:out0 type: asymmetric_affine
  618. D Process Conv_Conv_15_24 ...
  619. D RKNN output shape(convolution): (1 20 20 512)
  620. D Tensor @Conv_Conv_15_24:out0 type: asymmetric_affine
  621. D Process Relu_Relu_16_22 ...
  622. D RKNN output shape(relu): (1 20 20 512)
  623. D Tensor @Relu_Relu_16_22:out0 type: asymmetric_affine
  624. D Process Conv_Conv_17_20 ...
  625. D RKNN output shape(convolution): (1 21 21 512)
  626. D Tensor @Conv_Conv_17_20:out0 type: asymmetric_affine
  627. D Process Relu_Relu_18_18 ...
  628. D RKNN output shape(relu): (1 21 21 512)
  629. D Tensor @Relu_Relu_18_18:out0 type: asymmetric_affine
  630. D Process MaxPool_MaxPool_19_17 ...
  631. D RKNN output shape(pooling): (1 20 20 512)
  632. D Tensor @MaxPool_MaxPool_19_17:out0 type: asymmetric_affine
  633. D Process Conv_Conv_20_14 ...
  634. D RKNN output shape(convolution): (1 20 20 1024)
  635. D Tensor @Conv_Conv_20_14:out0 type: asymmetric_affine
  636. D Process Relu_Relu_21_11 ...
  637. D RKNN output shape(relu): (1 20 20 1024)
  638. D Tensor @Relu_Relu_21_11:out0 type: asymmetric_affine
  639. D Process Conv_Conv_22_8 ...
  640. D RKNN output shape(convolution): (1 20 20 256)
  641. D Tensor @Conv_Conv_22_8:out0 type: asymmetric_affine
  642. D Process Relu_Relu_23_6 ...
  643. D RKNN output shape(relu): (1 20 20 256)
  644. D Tensor @Relu_Relu_23_6:out0 type: asymmetric_affine
  645. D Process Conv_Conv_26_15 ...
  646. D RKNN output shape(convolution): (1 20 20 128)
  647. D Tensor @Conv_Conv_26_15:out0 type: asymmetric_affine
  648. D Process Relu_Relu_27_12 ...
  649. D RKNN output shape(relu): (1 20 20 128)
  650. D Tensor @Relu_Relu_27_12:out0 type: asymmetric_affine
  651. D Process Resize_Resize_29_9 ...
  652. D RKNN output shape(image_resize): (1 40 40 128)
  653. D Tensor @Resize_Resize_29_9:out0 type: asymmetric_affine
  654. D Process Concat_Concat_30_7 ...
  655. D RKNN output shape(concat): (1 40 40 384)
  656. D Tensor @Concat_Concat_30_7:out0 type: asymmetric_affine
  657. D Process Conv_Conv_31_5 ...
  658. D RKNN output shape(convolution): (1 40 40 256)
  659. D Tensor @Conv_Conv_31_5:out0 type: asymmetric_affine
  660. D Process Relu_Relu_32_3 ...
  661. D RKNN output shape(relu): (1 40 40 256)
  662. D Tensor @Relu_Relu_32_3:out0 type: asymmetric_affine
  663. D Process attach_Relu_Relu_32/out0_0 ...
  664. D RKNN output shape(output): (1 40 40 256)
  665. D Tensor @attach_Relu_Relu_32/out0_0:out0 type: asymmetric_affine
  666. D Process Conv_Conv_24_4 ...
  667. D RKNN output shape(convolution): (1 20 20 512)
  668. D Tensor @Conv_Conv_24_4:out0 type: asymmetric_affine
  669. D Process Relu_Relu_25_2 ...
  670. D RKNN output shape(relu): (1 20 20 512)
  671. D Tensor @Relu_Relu_25_2:out0 type: asymmetric_affine
  672. D Process attach_Relu_Relu_25/out0_1 ...
  673. D RKNN output shape(output): (1 20 20 512)
  674. D Tensor @attach_Relu_Relu_25/out0_1:out0 type: asymmetric_affine
  675. I Build torch-jit-export complete.
  676. I Initialzing network optimizer by /home/victor/.local/lib/python3.6/site-packages/rknn/config/npu_config/PUMA ...
  677. D Optimizing network with merge_ximum, qnt_adjust_coef, multiply_transform, add_extra_io, format_input_ops, auto_fill_zero_bias, conv_kernel_transform, twod_op_transform, strip_op, extend_unstack_split, merge_layer, transform_layer, broadcast_op, strip_op, auto_fill_reshape_zero, adjust_output_attrs, insert_dtype_converter
  678. I Conv_Conv_0_33 merged relu nexted.
  679. I Conv_Conv_3_30 merged relu nexted.
  680. I Conv_Conv_6_27 merged relu nexted.
  681. I Conv_Conv_9_21 merged relu nexted.
  682. I Conv_Conv_15_24 merged relu nexted.
  683. I Conv_Conv_17_20 merged relu nexted.
  684. I Conv_Conv_20_14 merged relu nexted.
  685. I Conv_Conv_22_8 merged relu nexted.
  686. I Conv_Conv_24_4 merged relu nexted.
  687. I Conv_Conv_26_15 merged relu nexted.
  688. I Conv_Conv_31_5 merged relu nexted.
  689. I Start T2C Switcher...
  690. D Optimizing network with broadcast_op, t2c_fc
  691. D convert Concat_Concat_30_7(concat) axis 3 to 1
  692. I End T2C Switcher...
  693. D Optimizing network with merge_ximum, qnt_adjust_coef, multiply_transform, add_extra_io, format_input_ops, auto_fill_zero_bias, conv_kernel_transform, twod_op_transform, strip_op, extend_unstack_split, merge_layer, transform_layer, broadcast_op, strip_op, auto_fill_reshape_zero, adjust_output_attrs, insert_dtype_converter
  694. I Conv_Conv_0_33 merged relu nexted.
  695. I Conv_Conv_3_30 merged relu nexted.
  696. I Conv_Conv_6_27 merged relu nexted.
  697. I Conv_Conv_9_21 merged relu nexted.
  698. I Conv_Conv_15_24 merged relu nexted.
  699. I Conv_Conv_17_20 merged relu nexted.
  700. I Conv_Conv_20_14 merged relu nexted.
  701. I Conv_Conv_22_8 merged relu nexted.
  702. I Conv_Conv_24_4 merged relu nexted.
  703. I Conv_Conv_26_15 merged relu nexted.
  704. I Conv_Conv_31_5 merged relu nexted.
  705. D Optimizing network with conv_1xn_transform, proposal_opt, c2drv_convert_axis, c2drv_convert_shape, c2drv_convert_array, c2drv_cast_dtype, c2drv_trans_data
  706. I Building data ...
  707. D /home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION

  708. D /home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/lib/x64_linux:/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/lib/x64_linux/RKm

  709. I Packing data ...
  710. D Packing Conv_Conv_0_33 ...
  711. D Quantize @Conv_Conv_0_33:bias to asymmetric_affine.
  712. D Quantize @Conv_Conv_0_33:weight to asymmetric_affine.
  713. D Packing Conv_Conv_12_13 ...
  714. D Quantize @Conv_Conv_12_13:bias to asymmetric_affine.
  715. D Quantize @Conv_Conv_12_13:weight to asymmetric_affine.
  716. D Packing Conv_Conv_15_24 ...
  717. D Quantize @Conv_Conv_15_24:bias to asymmetric_affine.
  718. D Quantize @Conv_Conv_15_24:weight to asymmetric_affine.
  719. D Packing Conv_Conv_17_20 ...
  720. D Quantize @Conv_Conv_17_20:bias to asymmetric_affine.
  721. D Quantize @Conv_Conv_17_20:weight to asymmetric_affine.
  722. D Packing Conv_Conv_20_14 ...
  723. D Quantize @Conv_Conv_20_14:bias to asymmetric_affine.
  724. D Quantize @Conv_Conv_20_14:weight to asymmetric_affine.
  725. D Packing Conv_Conv_22_8 ...
  726. D Quantize @Conv_Conv_22_8:bias to asymmetric_affine.
  727. D Quantize @Conv_Conv_22_8:weight to asymmetric_affine.
  728. D Packing Conv_Conv_24_4 ...
  729. D Quantize @Conv_Conv_24_4:bias to asymmetric_affine.
  730. D Quantize @Conv_Conv_24_4:weight to asymmetric_affine.
  731. D Packing Conv_Conv_26_15 ...
  732. D Quantize @Conv_Conv_26_15:bias to asymmetric_affine.
  733. D Quantize @Conv_Conv_26_15:weight to asymmetric_affine.
  734. D Packing Conv_Conv_31_5 ...
  735. D Quantize @Conv_Conv_31_5:bias to asymmetric_affine.
  736. D Quantize @Conv_Conv_31_5:weight to asymmetric_affine.
  737. D Packing Conv_Conv_3_30 ...
  738. D Quantize @Conv_Conv_3_30:bias to asymmetric_affine.
  739. D Quantize @Conv_Conv_3_30:weight to asymmetric_affine.
  740. D Packing Conv_Conv_6_27 ...
  741. D Quantize @Conv_Conv_6_27:bias to asymmetric_affine.
  742. D Quantize @Conv_Conv_6_27:weight to asymmetric_affine.
  743. D Packing Conv_Conv_9_21 ...
  744. D Quantize @Conv_Conv_9_21:bias to asymmetric_affine.
  745. D Quantize @Conv_Conv_9_21:weight to asymmetric_affine.
  746. D Generate fake input
  747. D Compiling start...
  748. D gcc -Wall -std=c++0x -I. -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/CL -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/VX -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ovxlib -D__linux__ -DLINUX -O3 -c rknn_torchjitexport.c
  749. D gcc -Wall -std=c++0x -I. -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/CL -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/VX -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ovxlib -D__linux__ -DLINUX -O3 -c rknn_pre_process.c
  750. D gcc -Wall -std=c++0x -I. -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/CL -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/VX -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ovxlib -D__linux__ -DLINUX -O3 -c main.c
  751. D gcc -Wall -std=c++0x -I. -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/CL -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/VX -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ovxlib -D__linux__ -DLINUX -O3 -c rknn_post_process.c
  752. D gcc -Wall -std=c++0x -I. -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/CL -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/VX -I/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/include/ovxlib -D__linux__ -DLINUX -O3 -O3  rknn_torchjitexport.o rknn_pre_process.o main.o rknn_post_process.o -L/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/lib -lOpenVX -lOpenVXU -lCLC -lVSC -lGAL -lovxlib -lEmulator -lvdtproxy -L/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/lib/RKm -lOpenVX -lOpenVXU -lCLC -lVSC -lGAL -lovxlib -lEmulator -lvdtproxy -L/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/lib/x64_linux -lOpenVX -lOpenVXU -lCLC -lVSC -lGAL -lovxlib -lEmulator -lvdtproxy -L/home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/lib/x64_linux/RKm -lOpenVX -lOpenVXU -lCLC -lVSC -lGAL -lovxlib -lEmulator -lvdtproxy /home/victor/.local/lib/python3.6/site-packages/rknn/simulator/LION/lib/x64_linux/libjpeg.a -o gen_nbg
  753. D Compiling finished.
  754. D Generate pre-compile model data start...
  755. D Create Neural Network: 91ms or 91039us
  756. D Verify...
  757. D Verify Graph: 2590ms or 2590676us
  758. D Start run graph [1] times...
  759. D Run the 1 time: 1.00ms or 1461.00us
  760. D vxProcessGraph execution time:
  761. D Total   1.00ms or 1466.00us
  762. D Average 1.47ms or 1466.00us
  763. D  --- Top5 ---
  764. D   0: 0.000000
  765. D   1: 0.000000
  766. D   2: 0.000000
  767. D   3: 0.000000
  768. D   4: 0.000000
  769. D Warning: NN VipSram is NULL, but SRAM is enabled!
  770. D Generate pre-compile model data finished.
  771. I Dump nbg input meta to /tmp/tmpqtwxevhd/nbg_unify/nbg_meta.json
  772. D output tensor id = 0, name = Relu_Relu_32/out0_0
  773. D output tensor id = 1, name = Relu_Relu_25/out0_1
  774. D input tensor id = 2, name = images_34
  775. I Build config finished.
复制代码





欢迎光临 Toybrick (https://t.rock-chips.com/) Powered by Discuz! X3.3