Convert onnx to tensorrt error on Jetson Xavier.

Robotics & Edge Computing Jetson & Embedded Systems Jetson AGX Xavier
1

Hello,

 I convert pytorch model to onnx successfully.

Error happened when run this command in terminal.

nvidia@jetson-0423718017743:/usr/src/tensorrt$ ./bin/trtexec --onnx=/home/nvidia/Projects/sys-trt-alphapose/data/alphapose/pose_model.onnx 
onnx: /home/nvidia/Projects/sys-trt-alphapose/data/alphapose/pose_model.onnx
----------------------------------------------------------------
Input filename:   /home/nvidia/Projects/sys-trt-alphapose/data/alphapose/pose_model.onnx
ONNX IR version:  0.0.3
Opset version:    6
Producer name:    pytorch
Producer version: 0.3
Domain:           
Model version:    0
Doc string:       
----------------------------------------------------------------
While parsing node number 12 [Shape -> "667"]:
ERROR: /home/erisuser/p4sw/sw/gpgpu/MachineLearning/DIT/release/5.0/parsers/onnxOpenSource/ModelImporter.cpp:116 In function importNode:
[8] No importer registered for op: Shape
failed to parse onnx file
Engine could not be created
Engine could not be created

onnx model baidu link:https://pan.baidu.com/s/1SxBspcIS81NnqY-lr58Jfg

How to solve this problem? Thank you in advance!

2

Hi,

The baidu link is not working.
And could you also share the log of converting onnx model with us?

Thanks.

3

Hi,

Do you use this Constant layer in your model? (Or just use Constant-1?)
https://github.com/onnx/onnx/blob/master/docs/Operators.md#Constant

If yes, could you share which type do you use?
Please noticed that some type is not supported by TensorRT currently.

Thanks.

4

Hi, Aastall

This my log of of converting onnx model.

[code]
chen@chen-System-Product-Name:~/object-detection/pytorch2onnx$ python3 sppe_pytorch2onnx.py
Loading pose model from ./models/sppe/duc_se.pth
graph(%actual_input_1 : Float(1, 3, 320, 256)
%learned_0 : Float(64, 3, 7, 7)
%learned_1 : Float(64)
%learned_2 : Float(64)
%learned_3 : Float(64)
%learned_4 : Float(64)
%learned_5 : Long()
%learned_6 : Float(64, 64, 1, 1)
%learned_7 : Float(64)
%learned_8 : Float(64)
%learned_9 : Float(64)
%learned_10 : Float(64)
%learned_11 : Long()
%learned_12 : Float(64, 64, 3, 3)
%learned_13 : Float(64)
%learned_14 : Float(64)
%learned_15 : Float(64)
%learned_16 : Float(64)
%learned_17 : Long()
%learned_18 : Float(256, 64, 1, 1)
%learned_19 : Float(256)
%learned_20 : Float(256)
%learned_21 : Float(256)
%learned_22 : Float(256)
%learned_23 : Long()
%learned_24 : Float(256, 256)
%learned_25 : Float(256)
%learned_26 : Float(256, 256)
%learned_27 : Float(256)
%learned_28 : Float(256, 64, 1, 1)
%learned_29 : Float(256)
%learned_30 : Float(256)
%learned_31 : Float(256)
%learned_32 : Float(256)
%learned_33 : Long()
%learned_34 : Float(64, 256, 1, 1)
%learned_35 : Float(64)
%learned_36 : Float(64)
%learned_37 : Float(64)
%learned_38 : Float(64)
%learned_39 : Long()
%learned_40 : Float(64, 64, 3, 3)
%learned_41 : Float(64)
%learned_42 : Float(64)
%learned_43 : Float(64)
%learned_44 : Float(64)
%learned_45 : Long()
%learned_46 : Float(256, 64, 1, 1)
%learned_47 : Float(256)
%learned_48 : Float(256)
%learned_49 : Float(256)
%learned_50 : Float(256)
%learned_51 : Long()
%learned_52 : Float(64, 256, 1, 1)
%learned_53 : Float(64)
%learned_54 : Float(64)
%learned_55 : Float(64)
%learned_56 : Float(64)
%learned_57 : Long()
%learned_58 : Float(64, 64, 3, 3)
%learned_59 : Float(64)
%learned_60 : Float(64)
%learned_61 : Float(64)
%learned_62 : Float(64)
%learned_63 : Long()
%learned_64 : Float(256, 64, 1, 1)
%learned_65 : Float(256)
%learned_66 : Float(256)
%learned_67 : Float(256)
%learned_68 : Float(256)
%learned_69 : Long()
%learned_70 : Float(128, 256, 1, 1)
%learned_71 : Float(128)
%learned_72 : Float(128)
%learned_73 : Float(128)
%learned_74 : Float(128)
%learned_75 : Long()
%learned_76 : Float(128, 128, 3, 3)
%learned_77 : Float(128)
%learned_78 : Float(128)
%learned_79 : Float(128)
%learned_80 : Float(128)
%learned_81 : Long()
%learned_82 : Float(512, 128, 1, 1)
%learned_83 : Float(512)
%learned_84 : Float(512)
%learned_85 : Float(512)
%learned_86 : Float(512)
%learned_87 : Long()
%learned_88 : Float(512, 512)
%learned_89 : Float(512)
%learned_90 : Float(512, 512)
%learned_91 : Float(512)
%learned_92 : Float(512, 256, 1, 1)
%learned_93 : Float(512)
%learned_94 : Float(512)
%learned_95 : Float(512)
%learned_96 : Float(512)
%learned_97 : Long()
%learned_98 : Float(128, 512, 1, 1)
%learned_99 : Float(128)
%learned_100 : Float(128)
%learned_101 : Float(128)
%learned_102 : Float(128)
%learned_103 : Long()
%learned_104 : Float(128, 128, 3, 3)
%learned_105 : Float(128)
%learned_106 : Float(128)
%learned_107 : Float(128)
%learned_108 : Float(128)
%learned_109 : Long()
%learned_110 : Float(512, 128, 1, 1)
%learned_111 : Float(512)
%learned_112 : Float(512)
%learned_113 : Float(512)
%learned_114 : Float(512)
%learned_115 : Long()
%learned_116 : Float(128, 512, 1, 1)
%learned_117 : Float(128)
%learned_118 : Float(128)
%learned_119 : Float(128)
%learned_120 : Float(128)
%learned_121 : Long()
%learned_122 : Float(128, 128, 3, 3)
%learned_123 : Float(128)
%learned_124 : Float(128)
%learned_125 : Float(128)
%learned_126 : Float(128)
%learned_127 : Long()
%learned_128 : Float(512, 128, 1, 1)
%learned_129 : Float(512)
%learned_130 : Float(512)
%learned_131 : Float(512)
%learned_132 : Float(512)
%learned_133 : Long()
%learned_134 : Float(128, 512, 1, 1)
%learned_135 : Float(128)
%learned_136 : Float(128)
%learned_137 : Float(128)
%learned_138 : Float(128)
%learned_139 : Long()
%learned_140 : Float(128, 128, 3, 3)
%learned_141 : Float(128)
%learned_142 : Float(128)
%learned_143 : Float(128)
%learned_144 : Float(128)
%learned_145 : Long()
%learned_146 : Float(512, 128, 1, 1)
%learned_147 : Float(512)
%learned_148 : Float(512)
%learned_149 : Float(512)
%learned_150 : Float(512)
%learned_151 : Long()
%learned_152 : Float(256, 512, 1, 1)
%learned_153 : Float(256)
%learned_154 : Float(256)
%learned_155 : Float(256)
%learned_156 : Float(256)
%learned_157 : Long()
%learned_158 : Float(256, 256, 3, 3)
%learned_159 : Float(256)
%learned_160 : Float(256)
%learned_161 : Float(256)
%learned_162 : Float(256)
%learned_163 : Long()
%learned_164 : Float(1024, 256, 1, 1)
%learned_165 : Float(1024)
%learned_166 : Float(1024)
%learned_167 : Float(1024)
%learned_168 : Float(1024)
%learned_169 : Long()
%learned_170 : Float(1024, 1024)
%learned_171 : Float(1024)
%learned_172 : Float(1024, 1024)
%learned_173 : Float(1024)
%learned_174 : Float(1024, 512, 1, 1)
%learned_175 : Float(1024)
%learned_176 : Float(1024)
%learned_177 : Float(1024)
%learned_178 : Float(1024)
%learned_179 : Long()
%learned_180 : Float(256, 1024, 1, 1)
%learned_181 : Float(256)
%learned_182 : Float(256)
%learned_183 : Float(256)
%learned_184 : Float(256)
%learned_185 : Long()
%learned_186 : Float(256, 256, 3, 3)
%learned_187 : Float(256)
%learned_188 : Float(256)
%learned_189 : Float(256)
%learned_190 : Float(256)
%learned_191 : Long()
%learned_192 : Float(1024, 256, 1, 1)
%learned_193 : Float(1024)
%learned_194 : Float(1024)
%learned_195 : Float(1024)
%learned_196 : Float(1024)
%learned_197 : Long()
%learned_198 : Float(256, 1024, 1, 1)
%learned_199 : Float(256)
%learned_200 : Float(256)
%learned_201 : Float(256)
%learned_202 : Float(256)
%learned_203 : Long()
%learned_204 : Float(256, 256, 3, 3)
%learned_205 : Float(256)
%learned_206 : Float(256)
%learned_207 : Float(256)
%learned_208 : Float(256)
%learned_209 : Long()
%learned_210 : Float(1024, 256, 1, 1)
%learned_211 : Float(1024)
%learned_212 : Float(1024)
%learned_213 : Float(1024)
%learned_214 : Float(1024)
%learned_215 : Long()
%learned_216 : Float(256, 1024, 1, 1)
%learned_217 : Float(256)
%learned_218 : Float(256)
%learned_219 : Float(256)
%learned_220 : Float(256)
%learned_221 : Long()
%learned_222 : Float(256, 256, 3, 3)
%learned_223 : Float(256)
%learned_224 : Float(256)
%learned_225 : Float(256)
%learned_226 : Float(256)
%learned_227 : Long()
%learned_228 : Float(1024, 256, 1, 1)
%learned_229 : Float(1024)
%learned_230 : Float(1024)
%learned_231 : Float(1024)
%learned_232 : Float(1024)
%learned_233 : Long()
%learned_234 : Float(256, 1024, 1, 1)
%learned_235 : Float(256)
%learned_236 : Float(256)
%learned_237 : Float(256)
%learned_238 : Float(256)
%learned_239 : Long()
%learned_240 : Float(256, 256, 3, 3)
%learned_241 : Float(256)
%learned_242 : Float(256)
%learned_243 : Float(256)
%learned_244 : Float(256)
%learned_245 : Long()
%learned_246 : Float(1024, 256, 1, 1)
%learned_247 : Float(1024)
%learned_248 : Float(1024)
%learned_249 : Float(1024)
%learned_250 : Float(1024)
%learned_251 : Long()
%learned_252 : Float(256, 1024, 1, 1)
%learned_253 : Float(256)
%learned_254 : Float(256)
%learned_255 : Float(256)
%learned_256 : Float(256)
%learned_257 : Long()
%learned_258 : Float(256, 256, 3, 3)
%learned_259 : Float(256)
%learned_260 : Float(256)
%learned_261 : Float(256)
%learned_262 : Float(256)
%learned_263 : Long()
%learned_264 : Float(1024, 256, 1, 1)
%learned_265 : Float(1024)
%learned_266 : Float(1024)
%learned_267 : Float(1024)
%learned_268 : Float(1024)
%learned_269 : Long()
%learned_270 : Float(256, 1024, 1, 1)
%learned_271 : Float(256)
%learned_272 : Float(256)
%learned_273 : Float(256)
%learned_274 : Float(256)
%learned_275 : Long()
%learned_276 : Float(256, 256, 3, 3)
%learned_277 : Float(256)
%learned_278 : Float(256)
%learned_279 : Float(256)
%learned_280 : Float(256)
%learned_281 : Long()
%learned_282 : Float(1024, 256, 1, 1)
%learned_283 : Float(1024)
%learned_284 : Float(1024)
%learned_285 : Float(1024)
%learned_286 : Float(1024)
%learned_287 : Long()
%learned_288 : Float(256, 1024, 1, 1)
%learned_289 : Float(256)
%learned_290 : Float(256)
%learned_291 : Float(256)
%learned_292 : Float(256)
%learned_293 : Long()
%learned_294 : Float(256, 256, 3, 3)
%learned_295 : Float(256)
%learned_296 : Float(256)
%learned_297 : Float(256)
%learned_298 : Float(256)
%learned_299 : Long()
%learned_300 : Float(1024, 256, 1, 1)
%learned_301 : Float(1024)
%learned_302 : Float(1024)
%learned_303 : Float(1024)
%learned_304 : Float(1024)
%learned_305 : Long()
%learned_306 : Float(256, 1024, 1, 1)
%learned_307 : Float(256)
%learned_308 : Float(256)
%learned_309 : Float(256)
%learned_310 : Float(256)
%learned_311 : Long()
%learned_312 : Float(256, 256, 3, 3)
%learned_313 : Float(256)
%learned_314 : Float(256)
%learned_315 : Float(256)
%learned_316 : Float(256)
%learned_317 : Long()
%learned_318 : Float(1024, 256, 1, 1)
%learned_319 : Float(1024)
%learned_320 : Float(1024)
%learned_321 : Float(1024)
%learned_322 : Float(1024)
%learned_323 : Long()
%learned_324 : Float(256, 1024, 1, 1)
%learned_325 : Float(256)
%learned_326 : Float(256)
%learned_327 : Float(256)
%learned_328 : Float(256)
%learned_329 : Long()
%learned_330 : Float(256, 256, 3, 3)
%learned_331 : Float(256)
%learned_332 : Float(256)
%learned_333 : Float(256)
%learned_334 : Float(256)
%learned_335 : Long()
%learned_336 : Float(1024, 256, 1, 1)
%learned_337 : Float(1024)
%learned_338 : Float(1024)
%learned_339 : Float(1024)
%learned_340 : Float(1024)
%learned_341 : Long()
%learned_342 : Float(256, 1024, 1, 1)
%learned_343 : Float(256)
%learned_344 : Float(256)
%learned_345 : Float(256)
%learned_346 : Float(256)
%learned_347 : Long()
%learned_348 : Float(256, 256, 3, 3)
%learned_349 : Float(256)
%learned_350 : Float(256)
%learned_351 : Float(256)
%learned_352 : Float(256)
%learned_353 : Long()
%learned_354 : Float(1024, 256, 1, 1)
%learned_355 : Float(1024)
%learned_356 : Float(1024)
%learned_357 : Float(1024)
%learned_358 : Float(1024)
%learned_359 : Long()
%learned_360 : Float(256, 1024, 1, 1)
%learned_361 : Float(256)
%learned_362 : Float(256)
%learned_363 : Float(256)
%learned_364 : Float(256)
%learned_365 : Long()
%learned_366 : Float(256, 256, 3, 3)
%learned_367 : Float(256)
%learned_368 : Float(256)
%learned_369 : Float(256)
%learned_370 : Float(256)
%learned_371 : Long()
%learned_372 : Float(1024, 256, 1, 1)
%learned_373 : Float(1024)
%learned_374 : Float(1024)
%learned_375 : Float(1024)
%learned_376 : Float(1024)
%learned_377 : Long()
%learned_378 : Float(256, 1024, 1, 1)
%learned_379 : Float(256)
%learned_380 : Float(256)
%learned_381 : Float(256)
%learned_382 : Float(256)
%learned_383 : Long()
%learned_384 : Float(256, 256, 3, 3)
%learned_385 : Float(256)
%learned_386 : Float(256)
%learned_387 : Float(256)
%learned_388 : Float(256)
%learned_389 : Long()
%learned_390 : Float(1024, 256, 1, 1)
%learned_391 : Float(1024)
%learned_392 : Float(1024)
%learned_393 : Float(1024)
%learned_394 : Float(1024)
%learned_395 : Long()
%learned_396 : Float(256, 1024, 1, 1)
%learned_397 : Float(256)
%learned_398 : Float(256)
%learned_399 : Float(256)
%learned_400 : Float(256)
%learned_401 : Long()
%learned_402 : Float(256, 256, 3, 3)
%learned_403 : Float(256)
%learned_404 : Float(256)
%learned_405 : Float(256)
%learned_406 : Float(256)
%learned_407 : Long()
%learned_408 : Float(1024, 256, 1, 1)
%learned_409 : Float(1024)
%learned_410 : Float(1024)
%learned_411 : Float(1024)
%learned_412 : Float(1024)
%learned_413 : Long()
%learned_414 : Float(256, 1024, 1, 1)
%learned_415 : Float(256)
%learned_416 : Float(256)
%learned_417 : Float(256)
%learned_418 : Float(256)
%learned_419 : Long()
%learned_420 : Float(256, 256, 3, 3)
%learned_421 : Float(256)
%learned_422 : Float(256)
%learned_423 : Float(256)
%learned_424 : Float(256)
%learned_425 : Long()
%learned_426 : Float(1024, 256, 1, 1)
%learned_427 : Float(1024)
%learned_428 : Float(1024)
%learned_429 : Float(1024)
%learned_430 : Float(1024)
%learned_431 : Long()
%learned_432 : Float(256, 1024, 1, 1)
%learned_433 : Float(256)
%learned_434 : Float(256)
%learned_435 : Float(256)
%learned_436 : Float(256)
%learned_437 : Long()
%learned_438 : Float(256, 256, 3, 3)
%learned_439 : Float(256)
%learned_440 : Float(256)
%learned_441 : Float(256)
%learned_442 : Float(256)
%learned_443 : Long()
%learned_444 : Float(1024, 256, 1, 1)
%learned_445 : Float(1024)
%learned_446 : Float(1024)
%learned_447 : Float(1024)
%learned_448 : Float(1024)
%learned_449 : Long()
%learned_450 : Float(256, 1024, 1, 1)
%learned_451 : Float(256)
%learned_452 : Float(256)
%learned_453 : Float(256)
%learned_454 : Float(256)
%learned_455 : Long()
%learned_456 : Float(256, 256, 3, 3)
%learned_457 : Float(256)
%learned_458 : Float(256)
%learned_459 : Float(256)
%learned_460 : Float(256)
%learned_461 : Long()
%learned_462 : Float(1024, 256, 1, 1)
%learned_463 : Float(1024)
%learned_464 : Float(1024)
%learned_465 : Float(1024)
%learned_466 : Float(1024)
%learned_467 : Long()
%learned_468 : Float(256, 1024, 1, 1)
%learned_469 : Float(256)
%learned_470 : Float(256)
%learned_471 : Float(256)
%learned_472 : Float(256)
%learned_473 : Long()
%learned_474 : Float(256, 256, 3, 3)
%learned_475 : Float(256)
%learned_476 : Float(256)
%learned_477 : Float(256)
%learned_478 : Float(256)
%learned_479 : Long()
%learned_480 : Float(1024, 256, 1, 1)
%learned_481 : Float(1024)
%learned_482 : Float(1024)
%learned_483 : Float(1024)
%learned_484 : Float(1024)
%learned_485 : Long()
%learned_486 : Float(256, 1024, 1, 1)
%learned_487 : Float(256)
%learned_488 : Float(256)
%learned_489 : Float(256)
%learned_490 : Float(256)
%learned_491 : Long()
%learned_492 : Float(256, 256, 3, 3)
%learned_493 : Float(256)
%learned_494 : Float(256)
%learned_495 : Float(256)
%learned_496 : Float(256)
%learned_497 : Long()
%learned_498 : Float(1024, 256, 1, 1)
%learned_499 : Float(1024)
%learned_500 : Float(1024)
%learned_501 : Float(1024)
%learned_502 : Float(1024)
%learned_503 : Long()
%learned_504 : Float(256, 1024, 1, 1)
%learned_505 : Float(256)
%learned_506 : Float(256)
%learned_507 : Float(256)
%learned_508 : Float(256)
%learned_509 : Long()
%learned_510 : Float(256, 256, 3, 3)
%learned_511 : Float(256)
%learned_512 : Float(256)
%learned_513 : Float(256)
%learned_514 : Float(256)
%learned_515 : Long()
%learned_516 : Float(1024, 256, 1, 1)
%learned_517 : Float(1024)
%learned_518 : Float(1024)
%learned_519 : Float(1024)
%learned_520 : Float(1024)
%learned_521 : Long()
%learned_522 : Float(256, 1024, 1, 1)
%learned_523 : Float(256)
%learned_524 : Float(256)
%learned_525 : Float(256)
%learned_526 : Float(256)
%learned_527 : Long()
%learned_528 : Float(256, 256, 3, 3)
%learned_529 : Float(256)
%learned_530 : Float(256)
%learned_531 : Float(256)
%learned_532 : Float(256)
%learned_533 : Long()
%learned_534 : Float(1024, 256, 1, 1)
%learned_535 : Float(1024)
%learned_536 : Float(1024)
%learned_537 : Float(1024)
%learned_538 : Float(1024)
%learned_539 : Long()
%learned_540 : Float(256, 1024, 1, 1)
%learned_541 : Float(256)
%learned_542 : Float(256)
%learned_543 : Float(256)
%learned_544 : Float(256)
%learned_545 : Long()
%learned_546 : Float(256, 256, 3, 3)
%learned_547 : Float(256)
%learned_548 : Float(256)
%learned_549 : Float(256)
%learned_550 : Float(256)
%learned_551 : Long()
%learned_552 : Float(1024, 256, 1, 1)
%learned_553 : Float(1024)
%learned_554 : Float(1024)
%learned_555 : Float(1024)
%learned_556 : Float(1024)
%learned_557 : Long()
%learned_558 : Float(256, 1024, 1, 1)
%learned_559 : Float(256)
%learned_560 : Float(256)
%learned_561 : Float(256)
%learned_562 : Float(256)
%learned_563 : Long()
%learned_564 : Float(256, 256, 3, 3)
%learned_565 : Float(256)
%learned_566 : Float(256)
%learned_567 : Float(256)
%learned_568 : Float(256)
%learned_569 : Long()
%learned_570 : Float(1024, 256, 1, 1)
%learned_571 : Float(1024)
%learned_572 : Float(1024)
%learned_573 : Float(1024)
%learned_574 : Float(1024)
%learned_575 : Long()
%learned_576 : Float(512, 1024, 1, 1)
%learned_577 : Float(512)
%learned_578 : Float(512)
%learned_579 : Float(512)
%learned_580 : Float(512)
%learned_581 : Long()
%learned_582 : Float(512, 512, 3, 3)
%learned_583 : Float(512)
%learned_584 : Float(512)
%learned_585 : Float(512)
%learned_586 : Float(512)
%learned_587 : Long()
%learned_588 : Float(2048, 512, 1, 1)
%learned_589 : Float(2048)
%learned_590 : Float(2048)
%learned_591 : Float(2048)
%learned_592 : Float(2048)
%learned_593 : Long()
%learned_594 : Float(2048, 2048)
%learned_595 : Float(2048)
%learned_596 : Float(2048, 2048)
%learned_597 : Float(2048)
%learned_598 : Float(2048, 1024, 1, 1)
%learned_599 : Float(2048)
%learned_600 : Float(2048)
%learned_601 : Float(2048)
%learned_602 : Float(2048)
%learned_603 : Long()
%learned_604 : Float(512, 2048, 1, 1)
%learned_605 : Float(512)
%learned_606 : Float(512)
%learned_607 : Float(512)
%learned_608 : Float(512)
%learned_609 : Long()
%learned_610 : Float(512, 512, 3, 3)
%learned_611 : Float(512)
%learned_612 : Float(512)
%learned_613 : Float(512)
%learned_614 : Float(512)
%learned_615 : Long()
%learned_616 : Float(2048, 512, 1, 1)
%learned_617 : Float(2048)
%learned_618 : Float(2048)
%learned_619 : Float(2048)
%learned_620 : Float(2048)
%learned_621 : Long()
%learned_622 : Float(512, 2048, 1, 1)
%learned_623 : Float(512)
%learned_624 : Float(512)
%learned_625 : Float(512)
%learned_626 : Float(512)
%learned_627 : Long()
%learned_628 : Float(512, 512, 3, 3)
%learned_629 : Float(512)
%learned_630 : Float(512)
%learned_631 : Float(512)
%learned_632 : Float(512)
%learned_633 : Long()
%learned_634 : Float(2048, 512, 1, 1)
%learned_635 : Float(2048)
%learned_636 : Float(2048)
%learned_637 : Float(2048)
%learned_638 : Float(2048)
%learned_639 : Long()
%learned_640 : Float(1024, 512, 3, 3)
%learned_641 : Float(1024)
%learned_642 : Float(1024)
%learned_643 : Float(1024)
%learned_644 : Float(1024)
%learned_645 : Long()
%learned_646 : Float(512, 256, 3, 3)
%learned_647 : Float(512)
%learned_648 : Float(512)
%learned_649 : Float(512)
%learned_650 : Float(512)
%learned_651 : Long()
%learned_652 : Float(33, 128, 3, 3)
%learned_653 : Float(33)) {
%655 : Float(1, 64, 160, 128) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[7, 7], pads=[3, 3, 3, 3], strides=[2, 2]](%actual_input_1, %learned_0), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Conv2d[conv1]
%656 : Float(1, 64, 160, 128) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%655, %learned_1, %learned_2, %learned_3, %learned_4), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/BatchNorm2d[bn1]
%657 : Float(1, 64, 160, 128) = onnx::Relu(%656), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/ReLU[relu]
%658 : Float(1, 64, 80, 64) = onnx::MaxPoolkernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/MaxPool2d[maxpool]
%659 : Float(1, 64, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%658, %learned_6), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/Conv2d[conv1]
%660 : Float(1, 64, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%659, %learned_7, %learned_8, %learned_9, %learned_10), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/BatchNorm2d[bn1]
%661 : Float(1, 64, 80, 64) = onnx::Relu(%660), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]
%662 : Float(1, 64, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%661, %learned_12), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/Conv2d[conv2]
%663 : Float(1, 64, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%662, %learned_13, %learned_14, %learned_15, %learned_16), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/BatchNorm2d[bn2]
%664 : Float(1, 64, 80, 64) = onnx::Relu(%663), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]
%665 : Float(1, 256, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%664, %learned_18), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/Conv2d[conv3]
%666 : Float(1, 256, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%665, %learned_19, %learned_20, %learned_21, %learned_22), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/BatchNorm2d[bn3]
%667 : Dynamic = onnx::Shape(%666), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%668 : Dynamic = onnx::Sliceaxes=[0], ends=[1], starts=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%669 : Long() = onnx::Squeezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%670 : Dynamic = onnx::Shape(%666), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%671 : Dynamic = onnx::Sliceaxes=[0], ends=[2], starts=[1], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%672 : Long() = onnx::Squeezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%673 : Dynamic = onnx::Padmode=constant, pads=[0, 0, 0, 0, 0, 0, 0, 0], value=0, scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]/AvgPool2d
%674 : Float(1, 256, 1, 1) = onnx::AveragePoolkernel_shape=[80, 64], pads=[0, 0, 0, 0], strides=[1, 1], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]/AvgPool2d
%675 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%676 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%677 : Dynamic = onnx::Concat[axis=0](%675, %676), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%678 : Float(1, 256) = onnx::Reshape(%674, %677), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%679 : Float(1, 256) = onnx::Gemm[alpha=1, beta=1, broadcast=1, transB=1](%678, %learned_24, %learned_25), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]/Sequential[fc]/Linear[0]
%680 : Float(1, 256) = onnx::Relu(%679), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]/Sequential[fc]/ReLU[1]
%681 : Float(1, 256) = onnx::Gemm[alpha=1, beta=1, broadcast=1, transB=1](%680, %learned_26, %learned_27), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]/Sequential[fc]/Linear[2]
%682 : Float(1, 256) = onnx::Sigmoid(%681), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]/Sequential[fc]/Sigmoid[3]
%683 : Long() = onnx::Constantvalue={1}, scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%684 : Long() = onnx::Constantvalue={1}, scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%685 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%686 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%687 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%688 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%689 : Dynamic = onnx::Concat[axis=0](%685, %686, %687, %688), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%690 : Float(1, 256, 1, 1) = onnx::Reshape(%682, %689), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%691 : Float(1, 256, 80, 64) = onnx::Mul[broadcast=1, axis=0](%666, %690), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/SELayer[se]
%692 : Float(1, 256, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%658, %learned_28), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/Sequential[downsample]/Conv2d[0]
%693 : Float(1, 256, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%692, %learned_29, %learned_30, %learned_31, %learned_32), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]/Sequential[downsample]/BatchNorm2d[1]
%694 : Float(1, 256, 80, 64) = onnx::Add(%691, %693), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]
%695 : Float(1, 256, 80, 64) = onnx::Relu(%694), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[0]
%696 : Float(1, 64, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%695, %learned_34), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]/Conv2d[conv1]
%697 : Float(1, 64, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%696, %learned_35, %learned_36, %learned_37, %learned_38), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]/BatchNorm2d[bn1]
%698 : Float(1, 64, 80, 64) = onnx::Relu(%697), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]
%699 : Float(1, 64, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%698, %learned_40), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]/Conv2d[conv2]
%700 : Float(1, 64, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%699, %learned_41, %learned_42, %learned_43, %learned_44), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]/BatchNorm2d[bn2]
%701 : Float(1, 64, 80, 64) = onnx::Relu(%700), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]
%702 : Float(1, 256, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%701, %learned_46), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]/Conv2d[conv3]
%703 : Float(1, 256, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%702, %learned_47, %learned_48, %learned_49, %learned_50), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]/BatchNorm2d[bn3]
%704 : Float(1, 256, 80, 64) = onnx::Add(%703, %695), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]
%705 : Float(1, 256, 80, 64) = onnx::Relu(%704), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[1]
%706 : Float(1, 64, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%705, %learned_52), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]/Conv2d[conv1]
%707 : Float(1, 64, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%706, %learned_53, %learned_54, %learned_55, %learned_56), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]/BatchNorm2d[bn1]
%708 : Float(1, 64, 80, 64) = onnx::Relu(%707), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]
%709 : Float(1, 64, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%708, %learned_58), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]/Conv2d[conv2]
%710 : Float(1, 64, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%709, %learned_59, %learned_60, %learned_61, %learned_62), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]/BatchNorm2d[bn2]
%711 : Float(1, 64, 80, 64) = onnx::Relu(%710), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]
%712 : Float(1, 256, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%711, %learned_64), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]/Conv2d[conv3]
%713 : Float(1, 256, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%712, %learned_65, %learned_66, %learned_67, %learned_68), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]/BatchNorm2d[bn3]
%714 : Float(1, 256, 80, 64) = onnx::Add(%713, %705), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]
%715 : Float(1, 256, 80, 64) = onnx::Relu(%714), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer1]/Bottleneck[2]
%716 : Float(1, 128, 80, 64) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%715, %learned_70), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/Conv2d[conv1]
%717 : Float(1, 128, 80, 64) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%716, %learned_71, %learned_72, %learned_73, %learned_74), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/BatchNorm2d[bn1]
%718 : Float(1, 128, 80, 64) = onnx::Relu(%717), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]
%719 : Float(1, 128, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%718, %learned_76), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/Conv2d[conv2]
%720 : Float(1, 128, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%719, %learned_77, %learned_78, %learned_79, %learned_80), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/BatchNorm2d[bn2]
%721 : Float(1, 128, 40, 32) = onnx::Relu(%720), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]
%722 : Float(1, 512, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%721, %learned_82), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/Conv2d[conv3]
%723 : Float(1, 512, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%722, %learned_83, %learned_84, %learned_85, %learned_86), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/BatchNorm2d[bn3]
%724 : Dynamic = onnx::Shape(%723), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%725 : Dynamic = onnx::Sliceaxes=[0], ends=[1], starts=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%726 : Long() = onnx::Squeezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%727 : Dynamic = onnx::Shape(%723), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%728 : Dynamic = onnx::Sliceaxes=[0], ends=[2], starts=[1], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%729 : Long() = onnx::Squeezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%730 : Dynamic = onnx::Padmode=constant, pads=[0, 0, 0, 0, 0, 0, 0, 0], value=0, scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]/AvgPool2d
%731 : Float(1, 512, 1, 1) = onnx::AveragePoolkernel_shape=[40, 32], pads=[0, 0, 0, 0], strides=[1, 1], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]/AvgPool2d
%732 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%733 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%734 : Dynamic = onnx::Concat[axis=0](%732, %733), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%735 : Float(1, 512) = onnx::Reshape(%731, %734), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%736 : Float(1, 512) = onnx::Gemm[alpha=1, beta=1, broadcast=1, transB=1](%735, %learned_88, %learned_89), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]/Sequential[fc]/Linear[0]
%737 : Float(1, 512) = onnx::Relu(%736), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]/Sequential[fc]/ReLU[1]
%738 : Float(1, 512) = onnx::Gemm[alpha=1, beta=1, broadcast=1, transB=1](%737, %learned_90, %learned_91), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]/Sequential[fc]/Linear[2]
%739 : Float(1, 512) = onnx::Sigmoid(%738), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]/Sequential[fc]/Sigmoid[3]
%740 : Long() = onnx::Constantvalue={1}, scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%741 : Long() = onnx::Constantvalue={1}, scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%742 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%743 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%744 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%745 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%746 : Dynamic = onnx::Concat[axis=0](%742, %743, %744, %745), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%747 : Float(1, 512, 1, 1) = onnx::Reshape(%739, %746), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%748 : Float(1, 512, 40, 32) = onnx::Mul[broadcast=1, axis=0](%723, %747), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/SELayer[se]
%749 : Float(1, 512, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[2, 2]](%715, %learned_92), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/Sequential[downsample]/Conv2d[0]
%750 : Float(1, 512, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%749, %learned_93, %learned_94, %learned_95, %learned_96), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]/Sequential[downsample]/BatchNorm2d[1]
%751 : Float(1, 512, 40, 32) = onnx::Add(%748, %750), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]
%752 : Float(1, 512, 40, 32) = onnx::Relu(%751), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[0]
%753 : Float(1, 128, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%752, %learned_98), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]/Conv2d[conv1]
%754 : Float(1, 128, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%753, %learned_99, %learned_100, %learned_101, %learned_102), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]/BatchNorm2d[bn1]
%755 : Float(1, 128, 40, 32) = onnx::Relu(%754), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]
%756 : Float(1, 128, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%755, %learned_104), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]/Conv2d[conv2]
%757 : Float(1, 128, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%756, %learned_105, %learned_106, %learned_107, %learned_108), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]/BatchNorm2d[bn2]
%758 : Float(1, 128, 40, 32) = onnx::Relu(%757), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]
%759 : Float(1, 512, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%758, %learned_110), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]/Conv2d[conv3]
%760 : Float(1, 512, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%759, %learned_111, %learned_112, %learned_113, %learned_114), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]/BatchNorm2d[bn3]
%761 : Float(1, 512, 40, 32) = onnx::Add(%760, %752), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]
%762 : Float(1, 512, 40, 32) = onnx::Relu(%761), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[1]
%763 : Float(1, 128, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%762, %learned_116), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]/Conv2d[conv1]
%764 : Float(1, 128, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%763, %learned_117, %learned_118, %learned_119, %learned_120), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]/BatchNorm2d[bn1]
%765 : Float(1, 128, 40, 32) = onnx::Relu(%764), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]
%766 : Float(1, 128, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%765, %learned_122), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]/Conv2d[conv2]
%767 : Float(1, 128, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%766, %learned_123, %learned_124, %learned_125, %learned_126), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]/BatchNorm2d[bn2]
%768 : Float(1, 128, 40, 32) = onnx::Relu(%767), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]
%769 : Float(1, 512, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%768, %learned_128), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]/Conv2d[conv3]
%770 : Float(1, 512, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%769, %learned_129, %learned_130, %learned_131, %learned_132), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]/BatchNorm2d[bn3]
%771 : Float(1, 512, 40, 32) = onnx::Add(%770, %762), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]
%772 : Float(1, 512, 40, 32) = onnx::Relu(%771), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[2]
%773 : Float(1, 128, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%772, %learned_134), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]/Conv2d[conv1]
%774 : Float(1, 128, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%773, %learned_135, %learned_136, %learned_137, %learned_138), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]/BatchNorm2d[bn1]
%775 : Float(1, 128, 40, 32) = onnx::Relu(%774), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]
%776 : Float(1, 128, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%775, %learned_140), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]/Conv2d[conv2]
%777 : Float(1, 128, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%776, %learned_141, %learned_142, %learned_143, %learned_144), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]/BatchNorm2d[bn2]
%778 : Float(1, 128, 40, 32) = onnx::Relu(%777), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]
%779 : Float(1, 512, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%778, %learned_146), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]/Conv2d[conv3]
%780 : Float(1, 512, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%779, %learned_147, %learned_148, %learned_149, %learned_150), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]/BatchNorm2d[bn3]
%781 : Float(1, 512, 40, 32) = onnx::Add(%780, %772), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]
%782 : Float(1, 512, 40, 32) = onnx::Relu(%781), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer2]/Bottleneck[3]
%783 : Float(1, 256, 40, 32) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%782, %learned_152), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/Conv2d[conv1]
%784 : Float(1, 256, 40, 32) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%783, %learned_153, %learned_154, %learned_155, %learned_156), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/BatchNorm2d[bn1]
%785 : Float(1, 256, 40, 32) = onnx::Relu(%784), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]
%786 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[2, 2]](%785, %learned_158), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/Conv2d[conv2]
%787 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%786, %learned_159, %learned_160, %learned_161, %learned_162), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/BatchNorm2d[bn2]
%788 : Float(1, 256, 20, 16) = onnx::Relu(%787), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]
%789 : Float(1, 1024, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%788, %learned_164), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/Conv2d[conv3]
%790 : Float(1, 1024, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%789, %learned_165, %learned_166, %learned_167, %learned_168), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/BatchNorm2d[bn3]
%791 : Dynamic = onnx::Shape(%790), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%792 : Dynamic = onnx::Sliceaxes=[0], ends=[1], starts=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%793 : Long() = onnx::Squeezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%794 : Dynamic = onnx::Shape(%790), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%795 : Dynamic = onnx::Sliceaxes=[0], ends=[2], starts=[1], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%796 : Long() = onnx::Squeezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%797 : Dynamic = onnx::Padmode=constant, pads=[0, 0, 0, 0, 0, 0, 0, 0], value=0, scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]/AvgPool2d
%798 : Float(1, 1024, 1, 1) = onnx::AveragePoolkernel_shape=[20, 16], pads=[0, 0, 0, 0], strides=[1, 1], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]/AvgPool2d
%799 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%800 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%801 : Dynamic = onnx::Concat[axis=0](%799, %800), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%802 : Float(1, 1024) = onnx::Reshape(%798, %801), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%803 : Float(1, 1024) = onnx::Gemm[alpha=1, beta=1, broadcast=1, transB=1](%802, %learned_170, %learned_171), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]/Sequential[fc]/Linear[0]
%804 : Float(1, 1024) = onnx::Relu(%803), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]/Sequential[fc]/ReLU[1]
%805 : Float(1, 1024) = onnx::Gemm[alpha=1, beta=1, broadcast=1, transB=1](%804, %learned_172, %learned_173), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]/Sequential[fc]/Linear[2]
%806 : Float(1, 1024) = onnx::Sigmoid(%805), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]/Sequential[fc]/Sigmoid[3]
%807 : Long() = onnx::Constantvalue={1}, scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%808 : Long() = onnx::Constantvalue={1}, scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%809 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%810 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%811 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%812 : Dynamic = onnx::Unsqueezeaxes=[0], scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%813 : Dynamic = onnx::Concat[axis=0](%809, %810, %811, %812), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%814 : Float(1, 1024, 1, 1) = onnx::Reshape(%806, %813), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%815 : Float(1, 1024, 20, 16) = onnx::Mul[broadcast=1, axis=0](%790, %814), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/SELayer[se]
%816 : Float(1, 1024, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[2, 2]](%782, %learned_174), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/Sequential[downsample]/Conv2d[0]
%817 : Float(1, 1024, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%816, %learned_175, %learned_176, %learned_177, %learned_178), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]/Sequential[downsample]/BatchNorm2d[1]
%818 : Float(1, 1024, 20, 16) = onnx::Add(%815, %817), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]
%819 : Float(1, 1024, 20, 16) = onnx::Relu(%818), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[0]
%820 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%819, %learned_180), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]/Conv2d[conv1]
%821 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%820, %learned_181, %learned_182, %learned_183, %learned_184), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]/BatchNorm2d[bn1]
%822 : Float(1, 256, 20, 16) = onnx::Relu(%821), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]
%823 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%822, %learned_186), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]/Conv2d[conv2]
%824 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%823, %learned_187, %learned_188, %learned_189, %learned_190), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]/BatchNorm2d[bn2]
%825 : Float(1, 256, 20, 16) = onnx::Relu(%824), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]
%826 : Float(1, 1024, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%825, %learned_192), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]/Conv2d[conv3]
%827 : Float(1, 1024, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%826, %learned_193, %learned_194, %learned_195, %learned_196), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]/BatchNorm2d[bn3]
%828 : Float(1, 1024, 20, 16) = onnx::Add(%827, %819), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]
%829 : Float(1, 1024, 20, 16) = onnx::Relu(%828), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[1]
%830 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%829, %learned_198), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]/Conv2d[conv1]
%831 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%830, %learned_199, %learned_200, %learned_201, %learned_202), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]/BatchNorm2d[bn1]
%832 : Float(1, 256, 20, 16) = onnx::Relu(%831), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]
%833 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%832, %learned_204), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]/Conv2d[conv2]
%834 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%833, %learned_205, %learned_206, %learned_207, %learned_208), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]/BatchNorm2d[bn2]
%835 : Float(1, 256, 20, 16) = onnx::Relu(%834), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]
%836 : Float(1, 1024, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%835, %learned_210), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]/Conv2d[conv3]
%837 : Float(1, 1024, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%836, %learned_211, %learned_212, %learned_213, %learned_214), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]/BatchNorm2d[bn3]
%838 : Float(1, 1024, 20, 16) = onnx::Add(%837, %829), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]
%839 : Float(1, 1024, 20, 16) = onnx::Relu(%838), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[2]
%840 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%839, %learned_216), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]/Conv2d[conv1]
%841 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%840, %learned_217, %learned_218, %learned_219, %learned_220), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]/BatchNorm2d[bn1]
%842 : Float(1, 256, 20, 16) = onnx::Relu(%841), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]
%843 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%842, %learned_222), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]/Conv2d[conv2]
%844 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%843, %learned_223, %learned_224, %learned_225, %learned_226), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]/BatchNorm2d[bn2]
%845 : Float(1, 256, 20, 16) = onnx::Relu(%844), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]
%846 : Float(1, 1024, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%845, %learned_228), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]/Conv2d[conv3]
%847 : Float(1, 1024, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%846, %learned_229, %learned_230, %learned_231, %learned_232), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]/BatchNorm2d[bn3]
%848 : Float(1, 1024, 20, 16) = onnx::Add(%847, %839), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]
%849 : Float(1, 1024, 20, 16) = onnx::Relu(%848), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[3]
%850 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%849, %learned_234), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]/Conv2d[conv1]
%851 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%850, %learned_235, %learned_236, %learned_237, %learned_238), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]/BatchNorm2d[bn1]
%852 : Float(1, 256, 20, 16) = onnx::Relu(%851), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]
%853 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%852, %learned_240), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]/Conv2d[conv2]
%854 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%853, %learned_241, %learned_242, %learned_243, %learned_244), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]/BatchNorm2d[bn2]
%855 : Float(1, 256, 20, 16) = onnx::Relu(%854), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]
%856 : Float(1, 1024, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%855, %learned_246), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]/Conv2d[conv3]
%857 : Float(1, 1024, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%856, %learned_247, %learned_248, %learned_249, %learned_250), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]/BatchNorm2d[bn3]
%858 : Float(1, 1024, 20, 16) = onnx::Add(%857, %849), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]
%859 : Float(1, 1024, 20, 16) = onnx::Relu(%858), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[4]
%860 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%859, %learned_252), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]/Conv2d[conv1]
%861 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%860, %learned_253, %learned_254, %learned_255, %learned_256), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]/BatchNorm2d[bn1]
%862 : Float(1, 256, 20, 16) = onnx::Relu(%861), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]
%863 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[3, 3], pads=[1, 1, 1, 1], strides=[1, 1]](%862, %learned_258), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]/Conv2d[conv2]
%864 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%863, %learned_259, %learned_260, %learned_261, %learned_262), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]/BatchNorm2d[bn2]
%865 : Float(1, 256, 20, 16) = onnx::Relu(%864), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]
%866 : Float(1, 1024, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%865, %learned_264), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]/Conv2d[conv3]
%867 : Float(1, 1024, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%866, %learned_265, %learned_266, %learned_267, %learned_268), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]/BatchNorm2d[bn3]
%868 : Float(1, 1024, 20, 16) = onnx::Add(%867, %859), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]
%869 : Float(1, 1024, 20, 16) = onnx::Relu(%868), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[5]
%870 : Float(1, 256, 20, 16) = onnx::Conv[dilations=[1, 1], group=1, kernel_shape=[1, 1], pads=[0, 0, 0, 0], strides=[1, 1]](%869, %learned_270), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[6]/Conv2d[conv1]
%871 : Float(1, 256, 20, 16) = onnx::BatchNormalization[epsilon=1e-05, is_test=1, momentum=1](%870, %learned_271, %learned_272, %learned_273, %learned_274), scope: InferenNet_fast/FastPose[pyranet]/SEResnet[preact]/Sequential[layer3]/Bottleneck[6]/BatchNorm2d[bn1]
%872 : Float(1, 256, 20, 16) = onnx::Relu(%871), s

5

Hi,

Is it possible to update the model architecture?
It will require you to do some training for the refinement.

I think the main issue is the long type in the constant layer.
TensorRT only supports INT8, FLOAT16, FLOAT32 currently.

Thanks.

6

Could you please explain how did you manage to transform alphapose to ONNX?
Could you please upload onnx model to something other than baidu?