pytorch->onnx->ncnn对mobileNetv3模型转换输出类别个数问题 #2517

sumingzhe123 · 2020-12-29T03:05:31Z

我在pytorch->onnx->ncnn对mobileNetv3模型转换的过程中，遇到了：1、onnx2ncnn步骤有不支持问题，使用了onnxsim对onnx模型进行了处理后，再进行onnx2ncnn时转换OK，得到解决；2、我在构建应用，使用extractor.extract()时需要blob的name或index，当我使用ncnn2mem将文件转换成*.id.h打开查看最后的index并使用后，出现输出为28个，而我需要输出的是4个啊，奇了怪了，咋回事呢？我又使用netron可视化工具分别查看了*.param和*.param.bin两个文件，发现*.param正常最后输出4个，.param.bin不正常最后输出的是28个，整体长度对比，.param要比*.param.bin的图要长一些，貌似好像被截断了，但我使用的是*.param啊，这到底是咋回事呢？

nihui · 2020-12-29T08:25:39Z

模型中所有的 blob 都可以通过 name 和 index 拿出来，id.h 中的 28 个 BLOB_XXX 一定包含了 param 中的全部 blob 名字了，如果你只用 param，无需使用 ncnn2mem 工具，直接用 param 和 bin 文件即可

https://github.com/Tencent/ncnn/wiki/use-ncnn-with-alexnet.zh#%E5%8A%A0%E8%BD%BD%E6%A8%A1%E5%9E%8B

sumingzhe123 · 2020-12-29T08:41:38Z

.param文件如下：
7767517
112 187
Input input.1 Convolution Conv_0 HardSwish Div_5 Split splitncnn_0 Convolution Conv_6 ReLU Relu_7 ConvolutionDepthWise Conv_8 ReLU Relu_9 Convolution Conv_10 BinaryOp Add_11 Convolution Conv_12 ReLU Relu_13 ConvolutionDepthWise Conv_14 ReLU Relu_15 Convolution Conv_16 Split splitncnn_1 Convolution Conv_17 ReLU Relu_18 ConvolutionDepthWise Conv_19 ReLU Relu_20 Convolution Conv_21 BinaryOp Add_22 Convolution Conv_23 ReLU Relu_24 ConvolutionDepthWise Conv_25 ReLU Relu_26 Convolution Conv_27 Split splitncnn_2 Pooling Convolution Conv_29 ReLU Relu_30 Convolution Conv_31 HardSigmoid Div_35 BinaryOp Mul_36 Split splitncnn_3 Convolution Conv_37 ReLU Relu_38 ConvolutionDepthWise Conv_39 ReLU Relu_40 Convolution Conv_41 Split splitncnn_4 Pooling Convolution Conv_43 ReLU Relu_44 Convolution Conv_45 HardSigmoid Div_49 BinaryOp Mul_50 BinaryOp Add_51 Split splitncnn_5 Convolution Conv_52 ReLU Relu_53 ConvolutionDepthWise Conv_54 ReLU Relu_55 Convolution Conv_56 Split splitncnn_6 Pooling Convolution Conv_58 ReLU Relu_59 Convolution Conv_60 HardSigmoid Div_64 BinaryOp Mul_65 BinaryOp Add_66 Convolution Conv_67 HardSwish Div_72 ConvolutionDepthWise Conv_73 HardSwish Div_78 Convolution Conv_79 Split splitncnn_7 Convolution Conv_80 HardSwish Div_85 ConvolutionDepthWise Conv_86 HardSwish Div_91 Convolution Conv_92 BinaryOp Add_93 Split splitncnn_8 Convolution Conv_94 HardSwish Div_99 ConvolutionDepthWise Conv_100 HardSwish Div_105 Convolution Conv_106 BinaryOp Add_107 Split splitncnn_9 Convolution Conv_108 HardSwish Div_113 ConvolutionDepthWise Conv_114 HardSwish Div_119 Convolution Conv_120 BinaryOp Add_121 Split splitncnn_10 Convolution Conv_122 HardSwish Div_127 ConvolutionDepthWise Conv_128 HardSwish Div_133 Convolution Conv_134 Split splitncnn_11 Pooling Convolution Conv_136 ReLU Relu_137 Convolution Conv_138 HardSigmoid Div_142 BinaryOp Mul_143 Convolution Conv_144 BinaryOp Add_145 Split splitncnn_12 Convolution Conv_146 HardSwish Div_151 ConvolutionDepthWise Conv_152 HardSwish Div_157 Convolution Conv_158 Split splitncnn_13 Pooling Convolution Conv_160 ReLU Relu_161 Convolution Conv_162 HardSigmoid Div_166 BinaryOp Mul_167 BinaryOp Add_168 Split splitncnn_14 Convolution Conv_169 HardSwish Div_174 ConvolutionDepthWise Conv_175 HardSwish Div_180 Convolution Conv_181 Split splitncnn_15 Pooling Convolution Conv_183 ReLU Relu_184 Convolution Conv_185 HardSigmoid Div_189 BinaryOp Mul_190 Convolution Conv_191 BinaryOp Add_192 Convolution Conv_193 HardSwish Div_198 ConvolutionDepthWise Conv_199 HardSwish Div_204 Convolution Conv_205 Split splitncnn_16 Pooling Convolution Conv_207 ReLU Relu_208 Convolution Conv_209 HardSigmoid Div_213 BinaryOp Mul_214 Split splitncnn_17 Convolution Conv_215 HardSwish Div_220 ConvolutionDepthWise Conv_221 HardSwish Div_226 Convolution Conv_227 Split splitncnn_18 Pooling Convolution Conv_229 ReLU Relu_230 Convolution Conv_231 HardSigmoid Div_235 BinaryOp Mul_236 BinaryOp Add_237 Convolution Conv_238 HardSwish Div_243 Pooling AveragePool_245 Reshape Reshape_251 InnerProduct Gemm_252 BatchNorm HardSwish Div_258 InnerProduct Gemm_259 经过ncnn2mem之后生成的.id.h如下：
#ifndef NCNN_INCLUDE_GUA #define NCNN_INCLUDE_GUA namespace onnx_MobileNet const int LAYER_input_1 = 0;
const int BLOB_input_1 = 0;
const int LAYER_Conv_0 = 1;
const int BLOB_785 = 1;
const int LAYER_Div_5 = 2;
const int BLOB_408 = 2;
const int LAYER_splitncnn_0 = 3;
const int BLOB_408_splitncnn_0 = 3;
const int BLOB_408_splitncnn_1 = 4;
const int LAYER_Conv_6 = 4;
const int BLOB_788 = 5;
const int LAYER_Relu_7 = 5;
const int BLOB_411 = 6;
const int LAYER_Conv_8 = 6;
const int BLOB_791 = 7;
const int LAYER_Relu_9 = 7;
const int BLOB_414 = 8;
const int LAYER_Conv_10 = 8;
const int BLOB_794 = 9;
const int LAYER_Add_11 = 9;
const int BLOB_417 = 10;
const int LAYER_Conv_12 = 10;
const int BLOB_797 = 11;
const int LAYER_Relu_13 = 11;
const int BLOB_420 = 12;
const int LAYER_Conv_14 = 12;
const int BLOB_800 = 13;
const int LAYER_Relu_15 = 13;
const int BLOB_423 = 14;
const int LAYER_Conv_16 = 14;
const int BLOB_803 = 15;
const int LAYER_splitncnn_1 = 15;
const int BLOB_803_splitncnn_0 const int BLOB_803_splitncnn_1 const int LAYER_Conv_17 = 16;
const int BLOB_806 = 18;
const int LAYER_Relu_18 = 17;
const int BLOB_428 = 19;
const int LAYER_Conv_19 = 18;
const int BLOB_809 = 20;
const int LAYER_Relu_20 = 19;
const int BLOB_431 = 21;
const int LAYER_Conv_21 = 20;
const int BLOB_812 = 22;
const int LAYER_Add_22 = 21;
const int BLOB_434 = 23;
const int LAYER_Conv_23 = 22;
const int BLOB_815 = 24;
const int LAYER_Relu_24 = 23;
const int BLOB_437 = 25;
const int LAYER_Conv_25 = 24;
const int BLOB_818 = 26;
const int LAYER_Relu_26 = 25;
const int BLOB_440 = 27;
const int LAYER_Conv_27 = 26;
const int BLOB_821 = 28;
const int LAYER_splitncnn_2 = 27;
const int BLOB_821_splitncnn_0 const int BLOB_821_splitncnn_1 const int LAYER_GlobalAveragePool_28 const int BLOB_443 = 31;
const int LAYER_Conv_29 = 29;
const int BLOB_824 = 32;
const int LAYER_Relu_30 = 30;
const int BLOB_446 = 33;
const int LAYER_Conv_31 = 31;
const int BLOB_827 = 34;
const int LAYER_Div_35 = 32;
const int BLOB_454 = 35;
const int LAYER_Mul_36 = 33;
const int BLOB_455 = 36;
const int LAYER_splitncnn_3 = 34;
const int BLOB_455_splitncnn_0 const int BLOB_455_splitncnn_1 const int LAYER_Conv_37 = 35;
const int BLOB_830 = 39;
const int LAYER_Relu_38 = 36;
const int BLOB_458 = 40;
const int LAYER_Conv_39 = 37;
const int BLOB_833 = 41;
const int LAYER_Relu_40 = 38;
const int BLOB_461 = 42;
const int LAYER_Conv_41 = 39;
const int BLOB_836 = 43;
const int LAYER_splitncnn_4 = 40;
const int BLOB_836_splitncnn_0 const int BLOB_836_splitncnn_1 const int LAYER_GlobalAveragePool_42 const int BLOB_464 = 46;
const int LAYER_Conv_43 = 42;
const int BLOB_839 = 47;
const int LAYER_Relu_44 = 43;
const int BLOB_467 = 48;
const int LAYER_Conv_45 = 44;
const int BLOB_842 = 49;
const int LAYER_Div_49 = 45;
const int BLOB_475 = 50;
const int LAYER_Mul_50 = 46;
const int BLOB_476 = 51;
const int LAYER_Add_51 = 47;
const int BLOB_477 = 52;
const int LAYER_splitncnn_5 = 48;
const int BLOB_477_splitncnn_0 const int BLOB_477_splitncnn_1 const int LAYER_Conv_52 = 49;
const int BLOB_845 = 55;
const int LAYER_Relu_53 = 50;
const int BLOB_480 = 56;
const int LAYER_Conv_54 = 51;
const int BLOB_848 = 57;
const int LAYER_Relu_55 = 52;
const int BLOB_483 = 58;
const int LAYER_Conv_56 = 53;
const int BLOB_851 = 59;
const int LAYER_splitncnn_6 = 54;
const int BLOB_851_splitncnn_0 const int BLOB_851_splitncnn_1 const int LAYER_GlobalAveragePool_57 const int BLOB_486 = 62;
const int LAYER_Conv_58 = 56;
const int BLOB_854 = 63;
const int LAYER_Relu_59 = 57;
const int BLOB_489 = 64;
const int LAYER_Conv_60 = 58;
const int BLOB_857 = 65;
const int LAYER_Div_64 = 59;
const int BLOB_497 = 66;
const int LAYER_Mul_65 = 60;
const int BLOB_498 = 67;
const int LAYER_Add_66 = 61;
const int BLOB_499 = 68;
const int LAYER_Conv_67 = 62;
const int BLOB_860 = 69;
const int LAYER_Div_72 = 63;
const int BLOB_508 = 70;
const int LAYER_Conv_73 = 64;
const int BLOB_863 = 71;
const int LAYER_Div_78 = 65;
const int BLOB_517 = 72;
const int LAYER_Conv_79 = 66;
const int BLOB_866 = 73;
const int LAYER_splitncnn_7 = 67;
const int BLOB_866_splitncnn_0 const int BLOB_866_splitncnn_1 const int LAYER_Conv_80 = 68;
const int BLOB_869 = 76;
const int LAYER_Div_85 = 69;
const int BLOB_528 = 77;
const int LAYER_Conv_86 = 70;
const int BLOB_872 = 78;
const int LAYER_Div_91 = 71;
const int BLOB_537 = 79;
const int LAYER_Conv_92 = 72;
const int BLOB_875 = 80;
const int LAYER_Add_93 = 73;
const int BLOB_540 = 81;
const int LAYER_splitncnn_8 = 74;
const int BLOB_540_splitncnn_0 const int BLOB_540_splitncnn_1 const int LAYER_Conv_94 = 75;
const int BLOB_878 = 84;
const int LAYER_Div_99 = 76;
const int BLOB_549 = 85;
const int LAYER_Conv_100 = 77;
const int BLOB_881 = 86;
const int LAYER_Div_105 = 78;
const int BLOB_558 = 87;
const int LAYER_Conv_106 = 79;
const int BLOB_884 = 88;
const int LAYER_Add_107 = 80;
const int BLOB_561 = 89;
const int LAYER_splitncnn_9 = 81;
const int BLOB_561_splitncnn_0 const int BLOB_561_splitncnn_1 const int LAYER_Conv_108 = 82;
const int BLOB_887 = 92;
const int LAYER_Div_113 = 83;
const int BLOB_570 = 93;
const int LAYER_Conv_114 = 84;
const int BLOB_890 = 94;
const int LAYER_Div_119 = 85;
const int BLOB_579 = 95;
const int LAYER_Conv_120 = 86;
const int BLOB_893 = 96;
const int LAYER_Add_121 = 87;
const int BLOB_582 = 97;
const int LAYER_splitncnn_10 = 88;
const int BLOB_582_splitncnn_0 const int BLOB_582_splitncnn_1 const int LAYER_Conv_122 = 89;
const int BLOB_896 = 100;
const int LAYER_Div_127 = 90;
const int BLOB_591 = 101;
const int LAYER_Conv_128 = 91;
const int BLOB_899 = 102;
const int LAYER_Div_133 = 92;
const int BLOB_600 = 103;
const int LAYER_Conv_134 = 93;
const int BLOB_902 = 104;
const int LAYER_splitncnn_11 = 94;
const int BLOB_902_splitncnn_0 const int BLOB_902_splitncnn_1 const int LAYER_GlobalAveragePool_135 const int BLOB_603 = 107;
const int LAYER_Conv_136 = 96;
const int BLOB_905 = 108;
const int LAYER_Relu_137 = 97;
const int BLOB_606 = 109;
const int LAYER_Conv_138 = 98;
const int BLOB_908 = 110;
const int LAYER_Div_142 = 99;
const int BLOB_614 = 111;
const int LAYER_Mul_143 = 100;
const int BLOB_615 = 112;
const int LAYER_Conv_144 = 101;
const int BLOB_911 = 113;
const int LAYER_Add_145 = 102;
const int BLOB_618 = 114;
const int LAYER_splitncnn_12 = 103;
const int BLOB_618_splitncnn_0 const int BLOB_618_splitncnn_1 const int LAYER_Conv_146 = 104;
const int BLOB_914 = 117;
const int LAYER_Div_151 = 105;
const int BLOB_627 = 118;
const int LAYER_Conv_152 = 106;
const int BLOB_917 = 119;
const int LAYER_Div_157 = 107;
const int BLOB_636 = 120;
const int LAYER_Conv_158 = 108;
const int BLOB_920 = 121;
const int LAYER_splitncnn_13 = 109;
const int BLOB_920_splitncnn_0 const int BLOB_920_splitncnn_1 const int LAYER_GlobalAveragePool_159 const int BLOB_639 = 124;
const int LAYER_Conv_160 = 111;
const int BLOB_923 = 125;
} // namespace #endif // NCNN_INCLUDE_G 我需要拿最后的*.p 0 1 input.1
1 1 input.1 785 0=16 1=3 11=3 2=1 12=1 3=2 13=2 4=1 14=1 15=1 16=1 5=1 6=432
1 1 785 408 0=1.666667e-01 1=5.000000e-01
1 2 408 408_splitncnn_0 408_splitncnn_1
1 1 408_splitncnn_1 788 0=16 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=256
1 1 788 411
1 1 411 791 0=16 1=3 11=3 2=1 12=1 3=1 13=1 4=1 14=1 15=1 16=1 5=1 6=144 7=16
1 1 791 414
1 1 414 794 0=16 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=256
2 1 794 408_splitncnn_0 417 0=0
1 1 417 797 0=64 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=1024
1 1 797 420
1 1 420 800 0=64 1=3 11=3 2=1 12=1 3=2 13=2 4=1 14=1 15=1 16=1 5=1 6=576 7=64
1 1 800 423
1 1 423 803 0=24 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=1536
1 2 803 803_splitncnn_0 803_splitncnn_1
1 1 803_splitncnn_1 806 0=72 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=1728
1 1 806 428
1 1 428 809 0=72 1=3 11=3 2=1 12=1 3=1 13=1 4=1 14=1 15=1 16=1 5=1 6=648 7=72
1 1 809 431
1 1 431 812 0=24 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=1728
2 1 812 803_splitncnn_0 434 0=0
1 1 434 815 0=72 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=1728
1 1 815 437
1 1 437 818 0=72 1=5 11=5 2=1 12=1 3=2 13=2 4=2 14=2 15=2 16=2 5=1 6=1800 7=72
1 1 818 440
1 1 440 821 0=40 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=2880
1 2 821 821_splitncnn_0 821_splitncnn_1
GlobalAveragePool_28 1 1 821_splitncnn_1 443 0=1 4=1
1 1 443 824 0=10 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=400
1 1 824 446
1 1 446 827 0=40 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=400
1 1 827 454 0=1.666667e-01 1=5.000000e-01
2 1 821_splitncnn_0 454 455 0=2
1 2 455 455_splitncnn_0 455_splitncnn_1
1 1 455_splitncnn_1 830 0=120 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=4800
1 1 830 458
1 1 458 833 0=120 1=5 11=5 2=1 12=1 3=1 13=1 4=2 14=2 15=2 16=2 5=1 6=3000 7=120
1 1 833 461
1 1 461 836 0=40 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=4800
1 2 836 836_splitncnn_0 836_splitncnn_1
GlobalAveragePool_42 1 1 836_splitncnn_1 464 0=1 4=1
1 1 464 839 0=10 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=400
1 1 839 467
1 1 467 842 0=40 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=400
1 1 842 475 0=1.666667e-01 1=5.000000e-01
2 1 836_splitncnn_0 475 476 0=2
2 1 476 455_splitncnn_0 477 0=0
1 2 477 477_splitncnn_0 477_splitncnn_1
1 1 477_splitncnn_1 845 0=120 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=4800
1 1 845 480
1 1 480 848 0=120 1=5 11=5 2=1 12=1 3=1 13=1 4=2 14=2 15=2 16=2 5=1 6=3000 7=120
1 1 848 483
1 1 483 851 0=40 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=4800
1 2 851 851_splitncnn_0 851_splitncnn_1
GlobalAveragePool_57 1 1 851_splitncnn_1 486 0=1 4=1
1 1 486 854 0=10 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=400
1 1 854 489
1 1 489 857 0=40 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=400
1 1 857 497 0=1.666667e-01 1=5.000000e-01
2 1 851_splitncnn_0 497 498 0=2
2 1 498 477_splitncnn_0 499 0=0
1 1 499 860 0=240 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=9600
1 1 860 508 0=1.666667e-01 1=5.000000e-01
1 1 508 863 0=240 1=3 11=3 2=1 12=1 3=2 13=2 4=1 14=1 15=1 16=1 5=1 6=2160 7=240
1 1 863 517 0=1.666667e-01 1=5.000000e-01
1 1 517 866 0=80 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=19200
1 2 866 866_splitncnn_0 866_splitncnn_1
1 1 866_splitncnn_1 869 0=200 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=16000
1 1 869 528 0=1.666667e-01 1=5.000000e-01
1 1 528 872 0=200 1=3 11=3 2=1 12=1 3=1 13=1 4=1 14=1 15=1 16=1 5=1 6=1800 7=200
1 1 872 537 0=1.666667e-01 1=5.000000e-01
1 1 537 875 0=80 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=16000
2 1 875 866_splitncnn_0 540 0=0
1 2 540 540_splitncnn_0 540_splitncnn_1
1 1 540_splitncnn_1 878 0=184 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=14720
1 1 878 549 0=1.666667e-01 1=5.000000e-01
1 1 549 881 0=184 1=3 11=3 2=1 12=1 3=1 13=1 4=1 14=1 15=1 16=1 5=1 6=1656 7=184
1 1 881 558 0=1.666667e-01 1=5.000000e-01
1 1 558 884 0=80 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=14720
2 1 884 540_splitncnn_0 561 0=0
1 2 561 561_splitncnn_0 561_splitncnn_1
1 1 561_splitncnn_1 887 0=184 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=14720
1 1 887 570 0=1.666667e-01 1=5.000000e-01
1 1 570 890 0=184 1=3 11=3 2=1 12=1 3=1 13=1 4=1 14=1 15=1 16=1 5=1 6=1656 7=184
1 1 890 579 0=1.666667e-01 1=5.000000e-01
1 1 579 893 0=80 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=14720
2 1 893 561_splitncnn_0 582 0=0
1 2 582 582_splitncnn_0 582_splitncnn_1
1 1 582_splitncnn_1 896 0=480 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=38400
1 1 896 591 0=1.666667e-01 1=5.000000e-01
1 1 591 899 0=480 1=3 11=3 2=1 12=1 3=1 13=1 4=1 14=1 15=1 16=1 5=1 6=4320 7=480
1 1 899 600 0=1.666667e-01 1=5.000000e-01
1 1 600 902 0=112 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=53760
1 2 902 902_splitncnn_0 902_splitncnn_1
GlobalAveragePool_135 1 1 902_splitncnn_1 603 0=1 4=1
1 1 603 905 0=28 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=3136
1 1 905 606
1 1 606 908 0=112 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=3136
1 1 908 614 0=1.666667e-01 1=5.000000e-01
2 1 902_splitncnn_0 614 615 0=2
1 1 582_splitncnn_0 911 0=112 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=8960
2 1 615 911 618 0=0
1 2 618 618_splitncnn_0 618_splitncnn_1
1 1 618_splitncnn_1 914 0=672 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=75264
1 1 914 627 0=1.666667e-01 1=5.000000e-01
1 1 627 917 0=672 1=3 11=3 2=1 12=1 3=1 13=1 4=1 14=1 15=1 16=1 5=1 6=6048 7=672
1 1 917 636 0=1.666667e-01 1=5.000000e-01
1 1 636 920 0=112 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=75264
1 2 920 920_splitncnn_0 920_splitncnn_1
GlobalAveragePool_159 1 1 920_splitncnn_1 639 0=1 4=1
1 1 639 923 0=28 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=3136
1 1 923 642
1 1 642 926 0=112 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=3136
1 1 926 650 0=1.666667e-01 1=5.000000e-01
2 1 920_splitncnn_0 650 651 0=2
2 1 651 618_splitncnn_0 652 0=0
1 2 652 652_splitncnn_0 652_splitncnn_1
1 1 652_splitncnn_1 929 0=672 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=75264
1 1 929 661 0=1.666667e-01 1=5.000000e-01
1 1 661 932 0=672 1=5 11=5 2=1 12=1 3=1 13=1 4=2 14=2 15=2 16=2 5=1 6=16800 7=672
1 1 932 670 0=1.666667e-01 1=5.000000e-01
1 1 670 935 0=160 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=107520
1 2 935 935_splitncnn_0 935_splitncnn_1
GlobalAveragePool_182 1 1 935_splitncnn_1 673 0=1 4=1
1 1 673 938 0=40 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=6400
1 1 938 676
1 1 676 941 0=160 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=6400
1 1 941 684 0=1.666667e-01 1=5.000000e-01
2 1 935_splitncnn_0 684 685 0=2
1 1 652_splitncnn_0 944 0=160 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=17920
2 1 685 944 688 0=0
1 1 688 947 0=672 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=107520
1 1 947 697 0=1.666667e-01 1=5.000000e-01
1 1 697 950 0=672 1=5 11=5 2=1 12=1 3=2 13=2 4=2 14=2 15=2 16=2 5=1 6=16800 7=672
1 1 950 706 0=1.666667e-01 1=5.000000e-01
1 1 706 953 0=160 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=107520
1 2 953 953_splitncnn_0 953_splitncnn_1
GlobalAveragePool_206 1 1 953_splitncnn_1 709 0=1 4=1
1 1 709 956 0=40 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=6400
1 1 956 712
1 1 712 959 0=160 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=6400
1 1 959 720 0=1.666667e-01 1=5.000000e-01
2 1 953_splitncnn_0 720 721 0=2
1 2 721 721_splitncnn_0 721_splitncnn_1
1 1 721_splitncnn_1 962 0=960 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=153600
1 1 962 730 0=1.666667e-01 1=5.000000e-01
1 1 730 965 0=960 1=5 11=5 2=1 12=1 3=1 13=1 4=2 14=2 15=2 16=2 5=1 6=24000 7=960
1 1 965 739 0=1.666667e-01 1=5.000000e-01
1 1 739 968 0=160 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=153600
1 2 968 968_splitncnn_0 968_splitncnn_1
GlobalAveragePool_228 1 1 968_splitncnn_1 742 0=1 4=1
1 1 742 971 0=40 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=6400
1 1 971 745
1 1 745 974 0=160 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=6400
1 1 974 753 0=1.666667e-01 1=5.000000e-01
2 1 968_splitncnn_0 753 754 0=2
2 1 754 721_splitncnn_0 755 0=0
1 1 755 977 0=960 1=1 11=1 2=1 12=1 3=1 13=1 4=0 14=0 15=0 16=0 5=1 6=153600
1 1 977 764 0=1.666667e-01 1=5.000000e-01
1 1 764 766 0=1 1=7 11=7 2=7 12=7 3=0 13=0 14=0 15=0 5=1 6=0
1 1 766 774 0=-1
1 1 774 775 0=1280 1=1 2=1228800
BatchNormalization_253 1 1 775 776 0=1280
1 1 776 783 0=1.666667e-01 1=5.000000e-01
1 1 783 784 0=4 1=1 2=5120
RD_onnx_MobileNetV3_Large_weather_id_h
RD_onnx_MobileNetV3_Large_weather_id_h
V3_Large_weather_param_id {
= 16;
= 17;
= 29;
= 30;
= 28;
= 37;
= 38;
= 44;
= 45;
= 41;
= 53;
= 54;
= 60;
= 61;
= 55;
= 74;
= 75;
= 82;
= 83;
= 90;
= 91;
= 98;
= 99;
= 105;
= 106;
= 95;
= 115;
= 116;
= 122;
= 123;
= 110;
onnx_MobileNetV3_Large_weather_param_id
UARD_onnx_MobileNetV3_Large_weather_id_h
aram中的Gemm_259，但是id.h中没有对应的；另外，我直接使用的是 param 和 bin 文件，但是在extract()中输入Gemm_259出现find_blob_index_by_name Gemm_259 failed的错误；（我是在windows中转换的）

nihui · 2020-12-29T08:51:18Z

Gemm_259 是 layer name，不是 blob name，Gemm_259 的输出 blob name 是 784
推荐用netron打开，点击Gemm_259这个op，右边也会显示 blob name

sumingzhe123 · 2020-12-29T08:54:38Z

使用的是 param 和 bin 文件，extractor.extract("784", output); 依旧出现find_blob_index_by_name 784 failed的错误

sumingzhe123 · 2020-12-29T09:06:56Z

mobilenet_v3.zip附上文件

nihui · 2020-12-29T09:09:49Z

使用的是 param 和 bin 文件，extractor.extract("784", output); 依旧出现find_blob_index_by_name 784 failed的错误

确认 load_param load_model 文件是否正确

sumingzhe123 · 2020-12-29T09:14:48Z

ncnn::Net mobileNet_v3;
mobileNet_v3.load_param("MobileNetV3_Larger.param");
mobileNet_v3.load_model("MobileNetV3_Large.bin");
ncnn::Extractor extractor = mobileNet_v3.create_extractor();
extractor.input(0, in);
ncnn::Mat output;
extractor.extract("784", output);
加载方式确认无误，前面上传附件，您可以验证分析一下

WangGeng0 · 2023-01-31T08:33:13Z

我也和你一样，遇到了同一个问题了，请问您解决了没？？？

nihui · 2024-08-05T07:28:05Z

针对onnx模型转换的各种问题，推荐使用最新的pnnx工具转换到ncnn
In view of various problems in onnx model conversion, it is recommended to use the latest pnnx tool to convert your model to ncnn

pip install pnnx
pnnx model.onnx inputshape=[1,3,224,224]

详细参考文档
Detailed reference documentation
https://github.com/pnnx/pnnx
https://github.com/Tencent/ncnn/wiki/use-ncnn-with-pytorch-or-onnx#how-to-use-pnnx

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

pytorch->onnx->ncnn对mobileNetv3模型转换输出类别个数问题 #2517

pytorch->onnx->ncnn对mobileNetv3模型转换输出类别个数问题 #2517

sumingzhe123 commented Dec 29, 2020

nihui commented Dec 29, 2020

sumingzhe123 commented Dec 29, 2020

nihui commented Dec 29, 2020

sumingzhe123 commented Dec 29, 2020

sumingzhe123 commented Dec 29, 2020

nihui commented Dec 29, 2020

sumingzhe123 commented Dec 29, 2020

WangGeng0 commented Jan 31, 2023

nihui commented Aug 5, 2024

pytorch->onnx->ncnn对mobileNetv3模型转换输出类别个数问题 #2517

pytorch->onnx->ncnn对mobileNetv3模型转换输出类别个数问题 #2517

Comments

sumingzhe123 commented Dec 29, 2020

nihui commented Dec 29, 2020

sumingzhe123 commented Dec 29, 2020

nihui commented Dec 29, 2020

sumingzhe123 commented Dec 29, 2020

sumingzhe123 commented Dec 29, 2020

nihui commented Dec 29, 2020

sumingzhe123 commented Dec 29, 2020

WangGeng0 commented Jan 31, 2023

nihui commented Aug 5, 2024