Repository navigation
Description
Question:
Hi!I cloned your code and used my own data to train a model. When I wanted to export the pt model file into an engine file for easy deployment to the JetsonNano, the following problems occurred:
code:
--python export --weights best.pt --include engine --device 0
log:
[TensorRT] INFO: [MemUsageChange] Init CUDA: CPU +147, GPU +5, now: CPU 2197, GPU 3785 (MiB)
[TensorRT] INFO: ----------------------------------------------------------------
[TensorRT] INFO: Input filename: weights/helmet.onnx
[TensorRT] INFO: ONNX IR version: 0.0.6
[TensorRT] INFO: Opset version: 13
[TensorRT] INFO: Producer name: pytorch
[TensorRT] INFO: Producer version: 1.8
[TensorRT] INFO: Domain:
[TensorRT] INFO: Model version: 0
[TensorRT] INFO: Doc string:
[TensorRT] INFO: ----------------------------------------------------------------
[TensorRT] WARNING: onnx2trt_utils.cpp:364: Your ONNX model has been generated with INT64 weights, while TensorRT does not natively support INT64. Attempting to cast down to INT32.
[TensorRT] WARNING: onnx2trt_utils.cpp:390: One or more weights outside the range of INT32 was clamped
[TensorRT] WARNING: onnx2trt_utils.cpp:390: One or more weights outside the range of INT32 was clamped
[TensorRT] WARNING: onnx2trt_utils.cpp:390: One or more weights outside the range of INT32 was clamped
TensorRT: Network Description:
TensorRT: input "images" with shape (1, 3, 640, 640) and dtype DataType.FLOAT
TensorRT: output "output" with shape (1, 25200, 7) and dtype DataType.FLOAT
TensorRT: output "475" with shape (1, 3, 80, 80, 7) and dtype DataType.FLOAT
TensorRT: output "541" with shape (1, 3, 40, 40, 7) and dtype DataType.FLOAT
TensorRT: output "607" with shape (1, 3, 20, 20, 7) and dtype DataType.FLOAT
TensorRT: building FP32 engine in weights/helmet_yolo_improve.engine
[TensorRT] INFO: [MemUsageSnapshot] Builder begin: CPU 2225 MiB, GPU 3795 MiB
[TensorRT] INFO: ---------- Layers Running on DLA ----------
[TensorRT] INFO: ---------- Layers Running on GPU ----------
[TensorRT] INFO: [GpuLayer] 291
[TensorRT] INFO: [GpuLayer] 293
[TensorRT] INFO: [GpuLayer] 337
[TensorRT] INFO: [GpuLayer] 339
[TensorRT] INFO: [GpuLayer] 383
[TensorRT] INFO: [GpuLayer] 385
[TensorRT] INFO: [GpuLayer] 424
[TensorRT] INFO: [GpuLayer] 426
[TensorRT] INFO: [GpuLayer] 486
[TensorRT] INFO: [GpuLayer] 499
[TensorRT] INFO: [GpuLayer] 552
[TensorRT] INFO: [GpuLayer] 565
[TensorRT] INFO: [GpuLayer] 618
[TensorRT] INFO: [GpuLayer] 631
[TensorRT] INFO: [GpuLayer] Conv_0
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_1, PWN(Tanh_2)), Mul_3)
[TensorRT] INFO: [GpuLayer] Conv_4
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_5, PWN(Tanh_6)), Mul_7)
[TensorRT] INFO: [GpuLayer] Conv_8 || Conv_21
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_9, PWN(Tanh_10)), Mul_11)
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_22, PWN(Tanh_23)), Mul_24)
[TensorRT] INFO: [GpuLayer] Conv_12
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_13, PWN(Tanh_14)), Mul_15)
[TensorRT] INFO: [GpuLayer] Conv_16
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(Softplus_17, PWN(Tanh_18)), Mul_19), Add_20)
[TensorRT] INFO: [GpuLayer] Conv_26
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_27, PWN(Tanh_28)), Mul_29)
[TensorRT] INFO: [GpuLayer] Conv_30
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_31, PWN(Tanh_32)), Mul_33)
[TensorRT] INFO: [GpuLayer] Conv_34 || Conv_56
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_35, PWN(Tanh_36)), Mul_37)
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_57, PWN(Tanh_58)), Mul_59)
[TensorRT] INFO: [GpuLayer] Conv_38
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_39, PWN(Tanh_40)), Mul_41)
[TensorRT] INFO: [GpuLayer] Conv_42
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(Softplus_43, PWN(Tanh_44)), Mul_45), Add_46)
[TensorRT] INFO: [GpuLayer] Conv_47
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_48, PWN(Tanh_49)), Mul_50)
[TensorRT] INFO: [GpuLayer] Conv_51
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(Softplus_52, PWN(Tanh_53)), Mul_54), Add_55)
[TensorRT] INFO: [GpuLayer] Conv_61
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_62, PWN(Tanh_63)), Mul_64)
[TensorRT] INFO: [GpuLayer] Conv_65
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_66, PWN(Tanh_67)), Mul_68)
[TensorRT] INFO: [GpuLayer] Conv_69 || Conv_100
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_70, PWN(Tanh_71)), Mul_72)
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_101, PWN(Tanh_102)), Mul_103)
[TensorRT] INFO: [GpuLayer] Conv_73
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_74, PWN(Tanh_75)), Mul_76)
[TensorRT] INFO: [GpuLayer] Conv_77
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(Softplus_78, PWN(Tanh_79)), Mul_80), Add_81)
[TensorRT] INFO: [GpuLayer] Conv_82
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_83, PWN(Tanh_84)), Mul_85)
[TensorRT] INFO: [GpuLayer] Conv_86
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(Softplus_87, PWN(Tanh_88)), Mul_89), Add_90)
[TensorRT] INFO: [GpuLayer] Conv_91
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_92, PWN(Tanh_93)), Mul_94)
[TensorRT] INFO: [GpuLayer] Conv_95
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(Softplus_96, PWN(Tanh_97)), Mul_98), Add_99)
[TensorRT] INFO: [GpuLayer] Conv_105
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_106, PWN(Tanh_107)), Mul_108)
[TensorRT] INFO: [GpuLayer] Conv_109
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_110, PWN(Tanh_111)), Mul_112)
[TensorRT] INFO: [GpuLayer] Conv_113 || Conv_126
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_114, PWN(Tanh_115)), Mul_116)
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_127, PWN(Tanh_128)), Mul_129)
[TensorRT] INFO: [GpuLayer] Conv_117
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_118, PWN(Tanh_119)), Mul_120)
[TensorRT] INFO: [GpuLayer] Conv_121
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(Softplus_122, PWN(Tanh_123)), Mul_124), Add_125)
[TensorRT] INFO: [GpuLayer] Conv_131
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_132, PWN(Tanh_133)), Mul_134)
[TensorRT] INFO: [GpuLayer] Conv_135
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_136, PWN(Tanh_137)), Mul_138)
[TensorRT] INFO: [GpuLayer] MaxPool_139
[TensorRT] INFO: [GpuLayer] MaxPool_140
[TensorRT] INFO: [GpuLayer] MaxPool_141
[TensorRT] INFO: [GpuLayer] 268 copy
[TensorRT] INFO: [GpuLayer] 269 copy
[TensorRT] INFO: [GpuLayer] 270 copy
[TensorRT] INFO: [GpuLayer] Conv_143
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_144, PWN(Tanh_145)), Mul_146)
[TensorRT] INFO: [GpuLayer] Conv_147
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_148, PWN(Tanh_149)), Mul_150)
[TensorRT] INFO: [GpuLayer] Conv_151
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_152, PWN(Tanh_153)), Mul_154)
[TensorRT] INFO: [GpuLayer] Reshape_157 + Transpose_158
[TensorRT] INFO: [GpuLayer] Reshape_160
[TensorRT] INFO: [GpuLayer] Gather_162
[TensorRT] INFO: [GpuLayer] Gather_164
[TensorRT] INFO: [GpuLayer] Resize_166
[TensorRT] INFO: [GpuLayer] 300 copy
[TensorRT] INFO: [GpuLayer] Conv_168 || Conv_180
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_169, PWN(Tanh_170)), Mul_171)
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_181, PWN(Tanh_182)), Mul_183)
[TensorRT] INFO: [GpuLayer] Conv_172
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_173, PWN(Tanh_174)), Mul_175)
[TensorRT] INFO: [GpuLayer] Conv_176
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_177, PWN(Tanh_178)), Mul_179)
[TensorRT] INFO: [GpuLayer] Conv_185
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_186, PWN(Tanh_187)), Mul_188)
[TensorRT] INFO: [GpuLayer] Conv_189
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_190, PWN(Tanh_191)), Mul_192)
[TensorRT] INFO: [GpuLayer] Conv_193
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_194, PWN(Tanh_195)), Mul_196)
[TensorRT] INFO: [GpuLayer] Reshape_199 + Transpose_200
[TensorRT] INFO: [GpuLayer] Reshape_202
[TensorRT] INFO: [GpuLayer] Gather_204
[TensorRT] INFO: [GpuLayer] Gather_206
[TensorRT] INFO: [GpuLayer] Resize_208
[TensorRT] INFO: [GpuLayer] 346 copy
[TensorRT] INFO: [GpuLayer] Conv_210 || Conv_222
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_211, PWN(Tanh_212)), Mul_213)
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_223, PWN(Tanh_224)), Mul_225)
[TensorRT] INFO: [GpuLayer] Conv_214
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_215, PWN(Tanh_216)), Mul_217)
[TensorRT] INFO: [GpuLayer] Conv_218
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_219, PWN(Tanh_220)), Mul_221)
[TensorRT] INFO: [GpuLayer] Conv_227
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_228, PWN(Tanh_229)), Mul_230)
[TensorRT] INFO: [GpuLayer] Conv_231
[TensorRT] INFO: [GpuLayer] Conv_313
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_232, PWN(Tanh_233)), Mul_234)
[TensorRT] INFO: [GpuLayer] Conv_235
[TensorRT] INFO: [GpuLayer] Reshape_330 + Transpose_331
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_236, PWN(Tanh_237)), Mul_238)
[TensorRT] INFO: [GpuLayer] PWN(Sigmoid_332)
[TensorRT] INFO: [GpuLayer] Slice_337
[TensorRT] INFO: [GpuLayer] Slice_350
[TensorRT] INFO: [GpuLayer] Slice_361
[TensorRT] INFO: [GpuLayer] Reshape_241 + Transpose_242
[TensorRT] INFO: [GpuLayer] Reshape_244
[TensorRT] INFO: [GpuLayer] PWN(PWN(495 + (Unnamed Layer* 338) [Shuffle] + Mul_352, PWN(497 + (Unnamed Layer* 341) [Shuffle], Pow_354)), Mul_356)
[TensorRT] INFO: [GpuLayer] Gather_246
[TensorRT] INFO: [GpuLayer] Gather_248
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(482 + (Unnamed Layer* 326) [Shuffle] + Mul_339, PWN(484 + (Unnamed Layer* 329) [Shuffle], Sub_341)), Add_343), 488 + (Unnamed Layer* 334) [Shuffle] + Mul_345)
[TensorRT] INFO: [GpuLayer] 489 copy
[TensorRT] INFO: [GpuLayer] 500 copy
[TensorRT] INFO: [GpuLayer] 505 copy
[TensorRT] INFO: [GpuLayer] 387 copy
[TensorRT] INFO: [GpuLayer] 341 copy
[TensorRT] INFO: [GpuLayer] Reshape_366
[TensorRT] INFO: [GpuLayer] Conv_251 || Conv_263
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_252, PWN(Tanh_253)), Mul_254)
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_264, PWN(Tanh_265)), Mul_266)
[TensorRT] INFO: [GpuLayer] Conv_255
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_256, PWN(Tanh_257)), Mul_258)
[TensorRT] INFO: [GpuLayer] Conv_259
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_260, PWN(Tanh_261)), Mul_262)
[TensorRT] INFO: [GpuLayer] Conv_268
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_269, PWN(Tanh_270)), Mul_271)
[TensorRT] INFO: [GpuLayer] Conv_272
[TensorRT] INFO: [GpuLayer] Conv_367
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_273, PWN(Tanh_274)), Mul_275)
[TensorRT] INFO: [GpuLayer] Conv_276
[TensorRT] INFO: [GpuLayer] Reshape_384 + Transpose_385
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_277, PWN(Tanh_278)), Mul_279)
[TensorRT] INFO: [GpuLayer] PWN(Sigmoid_386)
[TensorRT] INFO: [GpuLayer] Slice_391
[TensorRT] INFO: [GpuLayer] Slice_404
[TensorRT] INFO: [GpuLayer] Slice_415
[TensorRT] INFO: [GpuLayer] Reshape_282 + Transpose_283
[TensorRT] INFO: [GpuLayer] Reshape_285
[TensorRT] INFO: [GpuLayer] PWN(PWN(561 + (Unnamed Layer* 385) [Shuffle] + Mul_406, PWN(563 + (Unnamed Layer* 388) [Shuffle], Pow_408)), Mul_410)
[TensorRT] INFO: [GpuLayer] Gather_287
[TensorRT] INFO: [GpuLayer] Gather_289
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(548 + (Unnamed Layer* 373) [Shuffle] + Mul_393, PWN(550 + (Unnamed Layer* 376) [Shuffle], Sub_395)), Add_397), 554 + (Unnamed Layer* 381) [Shuffle] + Mul_399)
[TensorRT] INFO: [GpuLayer] 555 copy
[TensorRT] INFO: [GpuLayer] 566 copy
[TensorRT] INFO: [GpuLayer] 571 copy
[TensorRT] INFO: [GpuLayer] 428 copy
[TensorRT] INFO: [GpuLayer] 295 copy
[TensorRT] INFO: [GpuLayer] Reshape_420
[TensorRT] INFO: [GpuLayer] Conv_292 || Conv_304
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_293, PWN(Tanh_294)), Mul_295)
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_305, PWN(Tanh_306)), Mul_307)
[TensorRT] INFO: [GpuLayer] Conv_296
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_297, PWN(Tanh_298)), Mul_299)
[TensorRT] INFO: [GpuLayer] Conv_300
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_301, PWN(Tanh_302)), Mul_303)
[TensorRT] INFO: [GpuLayer] Conv_309
[TensorRT] INFO: [GpuLayer] PWN(PWN(Softplus_310, PWN(Tanh_311)), Mul_312)
[TensorRT] INFO: [GpuLayer] Conv_421
[TensorRT] INFO: [GpuLayer] Reshape_438 + Transpose_439
[TensorRT] INFO: [GpuLayer] PWN(Sigmoid_440)
[TensorRT] INFO: [GpuLayer] Slice_445
[TensorRT] INFO: [GpuLayer] Slice_458
[TensorRT] INFO: [GpuLayer] Slice_469
[TensorRT] INFO: [GpuLayer] PWN(PWN(627 + (Unnamed Layer* 432) [Shuffle] + Mul_460, PWN(629 + (Unnamed Layer* 435) [Shuffle], Pow_462)), Mul_464)
[TensorRT] INFO: [GpuLayer] PWN(PWN(PWN(614 + (Unnamed Layer* 420) [Shuffle] + Mul_447, PWN(616 + (Unnamed Layer* 423) [Shuffle], Sub_449)), Add_451), 620 + (Unnamed Layer* 428) [Shuffle] + Mul_453)
[TensorRT] INFO: [GpuLayer] 621 copy
[TensorRT] INFO: [GpuLayer] 632 copy
[TensorRT] INFO: [GpuLayer] 637 copy
[TensorRT] INFO: [GpuLayer] Reshape_474
[TensorRT] INFO: [GpuLayer] 516 copy
[TensorRT] INFO: [GpuLayer] 582 copy
[TensorRT] INFO: [GpuLayer] 648 copy
[TensorRT] INFO: [MemUsageChange] Init cuBLAS/cuBLASLt: CPU +0, GPU +2, now: CPU 2228, GPU 3797 (MiB)
[TensorRT] INFO: [MemUsageChange] Init cuDNN: CPU +1, GPU -3, now: CPU 2229, GPU 3794 (MiB)
[TensorRT] WARNING: Detected invalid timing cache, setup a local cache instead
[TensorRT] INFO: Some tactics do not have sufficient workspace memory to run. Increasing workspace size may increase performance, please check verbose output.
[TensorRT] INFO: Detected 1 inputs and 7 output network tensors.
[TensorRT] INFO: Total Host Persistent Memory: 131840
[TensorRT] INFO: Total Device Persistent Memory: 42272256
[TensorRT] INFO: Total Scratch Memory: 228864
[TensorRT] INFO: [MemUsageStats] Peak memory usage of TRT CPU/GPU memory allocators: CPU 7 MiB, GPU 96 MiB
[TensorRT] INFO: [MemUsageChange] Init cuBLAS/cuBLASLt: CPU +1, GPU +1, now: CPU 2239, GPU 3837 (MiB)
[TensorRT] INFO: [MemUsageChange] Init cuDNN: CPU +0, GPU +2, now: CPU 2239, GPU 3839 (MiB)
[TensorRT] INFO: [MemUsageChange] Init cuBLAS/cuBLASLt: CPU +0, GPU +0, now: CPU 2239, GPU 3839 (MiB)
[TensorRT] INFO: [MemUsageChange] Init cuBLAS/cuBLASLt: CPU +0, GPU +0, now: CPU 2238, GPU 3839 (MiB)
[TensorRT] INFO: [MemUsageSnapshot] Builder end: CPU 2238 MiB, GPU 3838 MiB
TensorRT: export failure: enter
Looking forward to any suggestions
Additional
OS:Ubuntu 18.04
python:3.6.7
tensorrt: 8.0.1.1
torch: 1.8.0
Activity
alanli1997 commentedon Nov 11, 2022
Hi, I have tested to export the GSConv-yolov5s-best.pt 2 the best.engine by the export.py, but I did not encounter errors only WARNING:
code:
--python export.py --weights runs/train/exp/weights/best.pt --include engine --device 0
log:
....
[11/11/2022-11:12:49] [TRT] [I] Total Activation Memory: 36454400
[11/11/2022-11:12:49] [TRT] [I] [MemUsageChange] Init cuDNN: CPU +0, GPU +8, now: CPU 3979, GPU 2199 (MiB)
[11/11/2022-11:12:49] [TRT] [W] TensorRT was linked against cuDNN 8.4.1 but loaded cuDNN 8.3.2
[11/11/2022-11:12:49] [TRT] [I] [MemUsageChange] TensorRT-managed allocation in building engine: CPU +0, GPU +0, now: CPU 0, GPU 0 (MiB)
[11/11/2022-11:12:49] [TRT] [W] The getMaxBatchSize() function should not be used with an engine built from a network created with NetworkDefinitionCreationFlag::kEXPLICIT_BATCH flag. This function will always return 1.
[11/11/2022-11:12:49] [TRT] [W] The getMaxBatchSize() function should not be used with an engine built from a network created with NetworkDefinitionCreationFlag::kEXPLICIT_BATCH flag. This function will always return 1.
TensorRT: export success, saved as runs/train/exp/weights/best.engine (26.9 MB)
Export complete (319.47s)
Results saved to runs/train/exp/weights
Detect: python detect.py --weights runs/train/exp/weights/best.engine
Validate: python val.py --weights runs/train/exp/weights/best.engine
PyTorch Hub: model = torch.hub.load('ultralytics/yolov5', 'custom', 'runs/train/exp/weights/best.engine')
Visualize: https://netron.app
depend
OS:Ubuntu 20.04
python:3.8
tensorrt: 8.4.3.1
torch: 1.12.1
So, if you wanna deploy the GSConv-yolov5 on the NANO by an earlier version of Python and TRT, you can try using the GSConvs or yolov5-6.0 (don‘t forget to add the GSConv module in your yolov5-6.0 project)
Goodluck🍺
alanli1997 commentedon Nov 11, 2022
Or use the GSConvs2 instead of the GSConv module.