graph torch-jit-export ( %0[FLOAT, 1x256x7x7] %1[FLOAT, 1x256x7x7] %2[FLOAT, 1x256x7x7] %3[FLOAT, 1x256x31x31] %4[FLOAT, 1x256x31x31] %5[FLOAT, 1x256x31x31] ) initializers ( %rpn2.cls.head.3.weight[FLOAT, 10x256x1x1] %rpn2.cls.head.3.bias[FLOAT, 10] %rpn2.loc.head.3.weight[FLOAT, 20x256x1x1] %rpn2.loc.head.3.bias[FLOAT, 20] %rpn3.cls.head.3.weight[FLOAT, 10x256x1x1] %rpn3.cls.head.3.bias[FLOAT, 10] %rpn3.loc.head.3.weight[FLOAT, 20x256x1x1] %rpn3.loc.head.3.bias[FLOAT, 20] %rpn4.cls.head.3.weight[FLOAT, 10x256x1x1] %rpn4.cls.head.3.bias[FLOAT, 10] %rpn4.loc.head.3.weight[FLOAT, 20x256x1x1] %rpn4.loc.head.3.bias[FLOAT, 20] %446[FLOAT, scalar] %450[FLOAT, scalar] %527[FLOAT, 256x256x3x3] %529[FLOAT, 256] %531[FLOAT, 256x256x3x3] %533[FLOAT, 256] %535[FLOAT, 256x256x1x1] %537[FLOAT, 256] %539[FLOAT, 256x256x3x3] %541[FLOAT, 256] %543[FLOAT, 256x256x3x3] %545[FLOAT, 256] %547[FLOAT, 256x256x1x1] %549[FLOAT, 256] %551[FLOAT, 256x256x3x3] %553[FLOAT, 256] %555[FLOAT, 256x256x3x3] %557[FLOAT, 256] %559[FLOAT, 256x256x1x1] %561[FLOAT, 256] %563[FLOAT, 256x256x3x3] %565[FLOAT, 256] %567[FLOAT, 256x256x3x3] %569[FLOAT, 256] %571[FLOAT, 256x256x1x1] %573[FLOAT, 256] %575[FLOAT, 256x256x3x3] %577[FLOAT, 256] %579[FLOAT, 256x256x3x3] %581[FLOAT, 256] %583[FLOAT, 256x256x1x1] %585[FLOAT, 256] %587[FLOAT, 256x256x3x3] %589[FLOAT, 256] %591[FLOAT, 256x256x3x3] %593[FLOAT, 256] %595[FLOAT, 256x256x1x1] %597[FLOAT, 256] %144[INT64, 4] %158[INT64, 4] %173[INT64, 4] %197[INT64, 4] %211[INT64, 4] %226[INT64, 4] %250[INT64, 4] %264[INT64, 4] %279[INT64, 4] %303[INT64, 4] %317[INT64, 4] %332[INT64, 4] %356[INT64, 4] %370[INT64, 4] %385[INT64, 4] %409[INT64, 4] %423[INT64, 4] %438[INT64, 4] ) { %126 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%0, %527, %529) %128 = Relu(%126) %129 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%3, %531, %533) %131 = Relu(%129) %145 = Reshape(%131, %144) %159 = Reshape(%128, %158) %160 = Conv[dilations = [1, 1], group = 256, kernel_shape = [5, 5], pads = [0, 0, 0, 0], strides = [1, 1]](%145, %159) %174 = Reshape(%160, %173) %175 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%174, %535, %537) %177 = Relu(%175) %178 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%177, %rpn2.cls.head.3.weight, %rpn2.cls.head.3.bias) %179 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%0, %539, %541) %181 = Relu(%179) %182 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%3, %543, %545) %184 = Relu(%182) %198 = Reshape(%184, %197) %212 = Reshape(%181, %211) %213 = Conv[dilations = [1, 1], group = 256, kernel_shape = [5, 5], pads = [0, 0, 0, 0], strides = [1, 1]](%198, %212) %227 = Reshape(%213, %226) %228 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%227, %547, %549) %230 = Relu(%228) %231 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%230, %rpn2.loc.head.3.weight, %rpn2.loc.head.3.bias) %232 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%1, %551, %553) %234 = Relu(%232) %235 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%4, %555, %557) %237 = Relu(%235) %251 = Reshape(%237, %250) %265 = Reshape(%234, %264) %266 = Conv[dilations = [1, 1], group = 256, kernel_shape = [5, 5], pads = [0, 0, 0, 0], strides = [1, 1]](%251, %265) %280 = Reshape(%266, %279) %281 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%280, %559, %561) %283 = Relu(%281) %284 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%283, %rpn3.cls.head.3.weight, %rpn3.cls.head.3.bias) %285 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%1, %563, %565) %287 = Relu(%285) %288 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%4, %567, %569) %290 = Relu(%288) %304 = Reshape(%290, %303) %318 = Reshape(%287, %317) %319 = Conv[dilations = [1, 1], group = 256, kernel_shape = [5, 5], pads = [0, 0, 0, 0], strides = [1, 1]](%304, %318) %333 = Reshape(%319, %332) %334 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%333, %571, %573) %336 = Relu(%334) %337 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%336, %rpn3.loc.head.3.weight, %rpn3.loc.head.3.bias) %338 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%2, %575, %577) %340 = Relu(%338) %341 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%5, %579, %581) %343 = Relu(%341) %357 = Reshape(%343, %356) %371 = Reshape(%340, %370) %372 = Conv[dilations = [1, 1], group = 256, kernel_shape = [5, 5], pads = [0, 0, 0, 0], strides = [1, 1]](%357, %371) %386 = Reshape(%372, %385) %387 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%386, %583, %585) %389 = Relu(%387) %390 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%389, %rpn4.cls.head.3.weight, %rpn4.cls.head.3.bias) %391 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%2, %587, %589) %393 = Relu(%391) %394 = Conv[dilations = [1, 1], group = 1, kernel_shape = [3, 3], pads = [0, 0, 0, 0], strides = [1, 1]](%5, %591, %593) %396 = Relu(%394) %410 = Reshape(%396, %409) %424 = Reshape(%393, %423) %425 = Conv[dilations = [1, 1], group = 256, kernel_shape = [5, 5], pads = [0, 0, 0, 0], strides = [1, 1]](%410, %424) %439 = Reshape(%425, %438) %440 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%439, %595, %597) %442 = Relu(%440) %443 = Conv[dilations = [1, 1], group = 1, kernel_shape = [1, 1], pads = [0, 0, 0, 0], strides = [1, 1]](%442, %rpn4.loc.head.3.weight, %rpn4.loc.head.3.bias) %444 = Add(%178, %284) %445 = Add(%444, %390) %447 = Div(%445, %446) %448 = Add(%231, %337) %449 = Add(%448, %443) %451 = Div(%449, %450) return %447, %451 }
the graph of the onnx is shown as above. I would like to know how to fix this problem.