NVIDIA / NVIDIA/TensorRT

Cask convolution isConsistent check failed

Open
#4,641 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Module:Engine Build
Dominant language
C++
Stars
13.4k
Forks
2.4k
Avg merge
5d 3h
Merged PRs (30d)
2

Description

Description

bridgedepth.py

我在使用trtexec的时候报错

[!TIP]
[11/11/2025-11:33:53] [V] [TRT] /bridge_model/depth_head/output_conv1/Conv (CaskConvolution[0x80000009]) profiling completed in 0.0160466 seconds. Fastest Tactic: 0x999e005e3b016ea6 Time: 0.338071
[11/11/2025-11:33:53] [V] [TRT] Skipping CaskFlattenConvolution: No valid tactics for /bridge_model/depth_head/output_conv1/Conv
[11/11/2025-11:33:53] [V] [TRT] >>>>>>>>>>>>>>> Chose Runner Type: CaskConvolution Tactic: 0x999e005e3b016ea6
[11/11/2025-11:33:53] [V] [TRT] =============== Computing costs for /bridge_model/depth_feature_down/Conv
[11/11/2025-11:33:53] [V] [TRT] *************** Autotuning format combination: Float((* 20480 E1),(* 160 E1),160,1) where E0=(- (MUL_ADD 16 (CEIL_DIV (+ height -15) 16) 16) height) E1=(CAST_F32_TO_I (FLOOR (MUL_ADD_F 2 (CAST_I_TO_F32 (MUL_ADD 4 (CEIL_DIV (MUL_ADD 7 (CEIL_DIV (+ (+ (MUL_ADD -16 (CEIL_DIV (+ E0 -15) 16) E0) height) -7) 8) -27) 14) 4)) 0))) -> Float((MUL_ADD 5120 E1 5120),(MUL_ADD 40 E1 40),40,1) where E0=(- (MUL_ADD 16 (CEIL_DIV (+ height -15) 16) 16) height) E1=(CEIL_DIV (+ (+ (MUL_ADD -16 (CEIL_DIV (+ E0 -15) 16) E0) height) -15) 8) where E0=(- (MUL_ADD 16 (CEIL_DIV (+ height -15) 16) 16) height) E1=(CAST_F32_TO_I (FLOOR (MUL_ADD_F 2 (CAST_I_TO_F32 (MUL_ADD 4 (CEIL_DIV (MUL_ADD 7 (CEIL_DIV (+ E2 -7) 8) -27) 14) 4)) 0))) E2=(+ (MUL_ADD -16 (CEIL_DIV (+ E0 -15) 16) E0) height) E3=(CEIL_DIV (+ E2 -15) 8) ***************
[11/11/2025-11:33:53] [E] Error[2]: [convBaseBuilder.cpp::nvinfer1::builder::CaskConvBaseBuilder<class nvinfer1::rt::task::CaskConvolutionRunner,-2147483639>::createConvolution::259] Error Code 2: Internal Error (Assertion isOpConsistent(convolution.get()) failed. Cask convolution isConsistent check failed.)

于是我是用polygraph进行了多轮尝试

(bridgedepth) E:\AI\Code\WorkCode\RT-Stereo\onnx>polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,320,320] right_image:[1,3,320,320]
[I] RUNNING | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,320,320] right_image:[1,3,320,320]
[I] Will generate inference input data according to provided TensorMetadata: {left_image [shape=(1, 3, 320, 320)],
     right_image [shape=(1, 3, 320, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:48:49  | Activating and starting inference
[I] Creating ONNX-Runtime Inference Session with providers: ['CPUExecutionProvider']
[I] onnxrt-runner-N0-11/11/25-14:48:49
    ---- Inference Input(s) ----
    {left_image [dtype=float32, shape=(1, 3, 320, 320)],
     right_image [dtype=float32, shape=(1, 3, 320, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:48:49
    ---- Inference Output(s) ----
    {disparity [dtype=float32, shape=(1, 320, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:48:49  | Completed 1 iteration(s) in 4672 ms | Average inference time: 4672 ms.
[I] PASSED | Runtime: 17.783s | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,320,320] right_image:[1,3,320,320]

(bridgedepth) E:\AI\Code\WorkCode\RT-Stereo\onnx>polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,256,320] right_image:[1,3,256,320]
[I] RUNNING | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,256,320] right_image:[1,3,256,320]
[I] Will generate inference input data according to provided TensorMetadata: {left_image [shape=(1, 3, 256, 320)],
     right_image [shape=(1, 3, 256, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:49:23  | Activating and starting inference
[I] Creating ONNX-Runtime Inference Session with providers: ['CPUExecutionProvider']
[I] onnxrt-runner-N0-11/11/25-14:49:23
    ---- Inference Input(s) ----
    {left_image [dtype=float32, shape=(1, 3, 256, 320)],
     right_image [dtype=float32, shape=(1, 3, 256, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:49:23
    ---- Inference Output(s) ----
    {disparity [dtype=float32, shape=(1, 256, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:49:23  | Completed 1 iteration(s) in 3535 ms | Average inference time: 3535 ms.
[I] PASSED | Runtime: 15.792s | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,256,320] right_image:[1,3,256,320]

(bridgedepth) E:\AI\Code\WorkCode\RT-Stereo\onnx>polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,128,320] right_image:[1,3,128,320]
[I] RUNNING | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,128,320] right_image:[1,3,128,320]
[I] Will generate inference input data according to provided TensorMetadata: {left_image [shape=(1, 3, 128, 320)],
     right_image [shape=(1, 3, 128, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:49:52  | Activating and starting inference
[I] Creating ONNX-Runtime Inference Session with providers: ['CPUExecutionProvider']
[I] onnxrt-runner-N0-11/11/25-14:49:52
    ---- Inference Input(s) ----
    {left_image [dtype=float32, shape=(1, 3, 128, 320)],
     right_image [dtype=float32, shape=(1, 3, 128, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:49:52
    ---- Inference Output(s) ----
    {disparity [dtype=float32, shape=(1, 128, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:49:52  | Completed 1 iteration(s) in 1923 ms | Average inference time: 1923 ms.
[I] PASSED | Runtime: 14.707s | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,128,320] right_image:[1,3,128,320]

(bridgedepth) E:\AI\Code\WorkCode\RT-Stereo\onnx>polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,32,320] right_image:[1,3,32,320]
[I] RUNNING | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,32,320] right_image:[1,3,32,320]
[I] Will generate inference input data according to provided TensorMetadata: {left_image [shape=(1, 3, 32, 320)],
     right_image [shape=(1, 3, 32, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:50:17  | Activating and starting inference
[I] Creating ONNX-Runtime Inference Session with providers: ['CPUExecutionProvider']
[I] onnxrt-runner-N0-11/11/25-14:50:17
    ---- Inference Input(s) ----
    {left_image [dtype=float32, shape=(1, 3, 32, 320)],
     right_image [dtype=float32, shape=(1, 3, 32, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:50:17
    ---- Inference Output(s) ----
    {disparity [dtype=float32, shape=(1, 32, 320)]}
[I] onnxrt-runner-N0-11/11/25-14:50:17  | Completed 1 iteration(s) in 720.6 ms | Average inference time: 720.6 ms.
[I] PASSED | Runtime: 13.987s | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --onnxrt --input-shapes left_image:[1,3,32,320] right_image:[1,3,32,320]

(bridgedepth) E:\AI\Code\WorkCode\RT-Stereo\onnx>polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,320,320] right_image:[1,3,320,320]
[I] RUNNING | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,320,320] right_image:[1,3,320,320]
[I] Will generate inference input data according to provided TensorMetadata: {left_image [shape=(1, 3, 320, 320)],
     right_image [shape=(1, 3, 320, 320)]}
[I] TF32 is disabled by default. Turn on TF32 for better performance with minor accuracy differences.
[I] trt-runner-N0-11/11/25-14:54:16     | Activating and starting inference
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:604] Reading dangerously large protocol message.  If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons.  To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:81] The total number of bytes read was 882710710
[I] Configuring with profiles:[
        Profile 0:
            {left_image [min=[1, 3, 320, 320], opt=[1, 3, 320, 320], max=[1, 3, 320, 320]],
             right_image [min=[1, 3, 320, 320], opt=[1, 3, 320, 320], max=[1, 3, 320, 320]]}
    ]
[W] profileSharing0806 is on by default in TensorRT 10.0. This flag is deprecated and has no effect.
[I] Building engine with configuration:
    Flags                  | []
    Engine Capability      | EngineCapability.STANDARD
    Memory Pools           | [WORKSPACE: 8187.50 MiB, TACTIC_DRAM: 8187.50 MiB, TACTIC_SHARED_MEMORY: 1024.00 MiB]
    Tactic Sources         | [EDGE_MASK_CONVOLUTIONS, JIT_CONVOLUTIONS]
    Profiling Verbosity    | ProfilingVerbosity.DETAILED
    Preview Features       | [PROFILE_SHARING_0806]
[I] Finished engine building in 114.424 seconds
[I] trt-runner-N0-11/11/25-14:54:16
    ---- Inference Input(s) ----
    {left_image [dtype=float32, shape=(1, 3, 320, 320)],
     right_image [dtype=float32, shape=(1, 3, 320, 320)]}
[I] trt-runner-N0-11/11/25-14:54:16
    ---- Inference Output(s) ----
    {disparity [dtype=float32, shape=(1, 320, 320)]}
[I] trt-runner-N0-11/11/25-14:54:16     | Completed 1 iteration(s) in 1508 ms | Average inference time: 1508 ms.
[I] PASSED | Runtime: 127.165s | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,320,320] right_image:[1,3,320,320]

(bridgedepth) E:\AI\Code\WorkCode\RT-Stereo\onnx>polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,256,320] right_image:[1,3,256,320]
[I] RUNNING | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,256,320] right_image:[1,3,256,320]
[I] Will generate inference input data according to provided TensorMetadata: {left_image [shape=(1, 3, 256, 320)],
     right_image [shape=(1, 3, 256, 320)]}
[I] TF32 is disabled by default. Turn on TF32 for better performance with minor accuracy differences.
[I] trt-runner-N0-11/11/25-14:57:02     | Activating and starting inference
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:604] Reading dangerously large protocol message.  If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons.  To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:81] The total number of bytes read was 882710710
[I] Configuring with profiles:[
        Profile 0:
            {left_image [min=[1, 3, 256, 320], opt=[1, 3, 256, 320], max=[1, 3, 256, 320]],
             right_image [min=[1, 3, 256, 320], opt=[1, 3, 256, 320], max=[1, 3, 256, 320]]}
    ]
[W] profileSharing0806 is on by default in TensorRT 10.0. This flag is deprecated and has no effect.
[I] Building engine with configuration:
    Flags                  | []
    Engine Capability      | EngineCapability.STANDARD
    Memory Pools           | [WORKSPACE: 8187.50 MiB, TACTIC_DRAM: 8187.50 MiB, TACTIC_SHARED_MEMORY: 1024.00 MiB]
    Tactic Sources         | [EDGE_MASK_CONVOLUTIONS, JIT_CONVOLUTIONS]
    Profiling Verbosity    | ProfilingVerbosity.DETAILED
    Preview Features       | [PROFILE_SHARING_0806]
[I] Finished engine building in 135.344 seconds
[I] trt-runner-N0-11/11/25-14:57:02
    ---- Inference Input(s) ----
    {left_image [dtype=float32, shape=(1, 3, 256, 320)],
     right_image [dtype=float32, shape=(1, 3, 256, 320)]}
[I] trt-runner-N0-11/11/25-14:57:02
    ---- Inference Output(s) ----
    {disparity [dtype=float32, shape=(1, 256, 320)]}
[I] trt-runner-N0-11/11/25-14:57:02     | Completed 1 iteration(s) in 1436 ms | Average inference time: 1436 ms.
[I] PASSED | Runtime: 147.875s | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,256,320] right_image:[1,3,256,320]

(bridgedepth) E:\AI\Code\WorkCode\RT-Stereo\onnx>polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,128,320] right_image:[1,3,128,320]
[I] RUNNING | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,128,320] right_image:[1,3,128,320]
[I] Will generate inference input data according to provided TensorMetadata: {left_image [shape=(1, 3, 128, 320)],
     right_image [shape=(1, 3, 128, 320)]}
[I] TF32 is disabled by default. Turn on TF32 for better performance with minor accuracy differences.
[I] trt-runner-N0-11/11/25-15:00:04     | Activating and starting inference
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:604] Reading dangerously large protocol message.  If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons.  To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:81] The total number of bytes read was 882710710
[I] Configuring with profiles:[
        Profile 0:
            {left_image [min=[1, 3, 128, 320], opt=[1, 3, 128, 320], max=[1, 3, 128, 320]],
             right_image [min=[1, 3, 128, 320], opt=[1, 3, 128, 320], max=[1, 3, 128, 320]]}
    ]
[W] profileSharing0806 is on by default in TensorRT 10.0. This flag is deprecated and has no effect.
[I] Building engine with configuration:
    Flags                  | []
    Engine Capability      | EngineCapability.STANDARD
    Memory Pools           | [WORKSPACE: 8187.50 MiB, TACTIC_DRAM: 8187.50 MiB, TACTIC_SHARED_MEMORY: 1024.00 MiB]
    Tactic Sources         | [EDGE_MASK_CONVOLUTIONS, JIT_CONVOLUTIONS]
    Profiling Verbosity    | ProfilingVerbosity.DETAILED
    Preview Features       | [PROFILE_SHARING_0806]
[I] Finished engine building in 116.202 seconds
[I] trt-runner-N0-11/11/25-15:00:04
    ---- Inference Input(s) ----
    {left_image [dtype=float32, shape=(1, 3, 128, 320)],
     right_image [dtype=float32, shape=(1, 3, 128, 320)]}
[I] trt-runner-N0-11/11/25-15:00:04
    ---- Inference Output(s) ----
    {disparity [dtype=float32, shape=(1, 128, 320)]}
[I] trt-runner-N0-11/11/25-15:00:04     | Completed 1 iteration(s) in 1450 ms | Average inference time: 1450 ms.
[I] PASSED | Runtime: 128.698s | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,128,320] right_image:[1,3,128,320]

(bridgedepth) E:\AI\Code\WorkCode\RT-Stereo\onnx>polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,32,320] right_image:[1,3,32,320]
[I] RUNNING | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,32,320] right_image:[1,3,32,320]
[I] Will generate inference input data according to provided TensorMetadata: {left_image [shape=(1, 3, 32, 320)],
     right_image [shape=(1, 3, 32, 320)]}
[I] TF32 is disabled by default. Turn on TF32 for better performance with minor accuracy differences.
[I] trt-runner-N0-11/11/25-15:02:42     | Activating and starting inference
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:604] Reading dangerously large protocol message.  If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons.  To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:81] The total number of bytes read was 882710710
[I] Configuring with profiles:[
        Profile 0:
            {left_image [min=[1, 3, 32, 320], opt=[1, 3, 32, 320], max=[1, 3, 32, 320]],
             right_image [min=[1, 3, 32, 320], opt=[1, 3, 32, 320], max=[1, 3, 32, 320]]}
    ]
[W] profileSharing0806 is on by default in TensorRT 10.0. This flag is deprecated and has no effect.
[I] Building engine with configuration:
    Flags                  | []
    Engine Capability      | EngineCapability.STANDARD
    Memory Pools           | [WORKSPACE: 8187.50 MiB, TACTIC_DRAM: 8187.50 MiB, TACTIC_SHARED_MEMORY: 1024.00 MiB]
    Tactic Sources         | [EDGE_MASK_CONVOLUTIONS, JIT_CONVOLUTIONS]
    Profiling Verbosity    | ProfilingVerbosity.DETAILED
    Preview Features       | [PROFILE_SHARING_0806]
[I] Finished engine building in 123.805 seconds
[I] trt-runner-N0-11/11/25-15:02:42
    ---- Inference Input(s) ----
    {left_image [dtype=float32, shape=(1, 3, 32, 320)],
     right_image [dtype=float32, shape=(1, 3, 32, 320)]}
[I] trt-runner-N0-11/11/25-15:02:42
    ---- Inference Output(s) ----
    {disparity [dtype=float32, shape=(1, 32, 320)]}
[I] trt-runner-N0-11/11/25-15:02:42     | Completed 1 iteration(s) in 1328 ms | Average inference time: 1328 ms.
[I] PASSED | Runtime: 136.176s | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --input-shapes left_image:[1,3,32,320] right_image:[1,3,32,320]

所有的静态都通过了，到动态就报错了

(bridgedepth) E:\AI\Code\WorkCode\RT-Stereo\onnx>polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --trt-min-shapes left_image:[1,3,32,320] right_image:[1,3,32,320] --trt-opt-shapes left_image:[1,3,320,320] right_image:[1,3,320,320] --trt-max-shapes left_image:[1,3,480,640] right_image:[1,3,480,640]
[I] RUNNING | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --trt-min-shapes left_image:[1,3,32,320] right_image:[1,3,32,320] --trt-opt-shapes left_image:[1,3,320,320] right_image:[1,3,320,320] --trt-max-shapes left_image:[1,3,480,640] right_image:[1,3,480,640]
[I] TF32 is disabled by default. Turn on TF32 for better performance with minor accuracy differences.
[I] trt-runner-N0-11/11/25-15:09:30     | Activating and starting inference
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:604] Reading dangerously large protocol message.  If the message turns out to be larger than 2147483647 bytes, parsing will be halted for security reasons.  To increase the limit (or to disable these warnings), see CodedInputStream::SetTotalBytesLimit() in google/protobuf/io/coded_stream.h.
[libprotobuf WARNING **************************************************************************\externals\protobuf\3.0.0\src\google\protobuf\io\coded_stream.cc:81] The total number of bytes read was 882710710
[I] Configuring with profiles:[
        Profile 0:
            {left_image [min=[1, 3, 32, 320], opt=[1, 3, 320, 320], max=[1, 3, 480, 640]],
             right_image [min=[1, 3, 32, 320], opt=[1, 3, 320, 320], max=[1, 3, 480, 640]]}
    ]
[W] profileSharing0806 is on by default in TensorRT 10.0. This flag is deprecated and has no effect.
[I] Building engine with configuration:
    Flags                  | []
    Engine Capability      | EngineCapability.STANDARD
    Memory Pools           | [WORKSPACE: 8187.50 MiB, TACTIC_DRAM: 8187.50 MiB, TACTIC_SHARED_MEMORY: 1024.00 MiB]
    Tactic Sources         | [EDGE_MASK_CONVOLUTIONS, JIT_CONVOLUTIONS]
    Profiling Verbosity    | ProfilingVerbosity.DETAILED
    Preview Features       | [PROFILE_SHARING_0806]
[E] [convBaseBuilder.cpp::nvinfer1::builder::CaskConvBaseBuilder<class nvinfer1::rt::task::CaskConvolutionRunner,-2147483639>::createConvolution::259] Error Code 2: Internal Error (Assertion isOpConsistent(convolution.get()) failed. Cask convolution isConsistent check failed.)
[!] Invalid Engine. Please ensure the engine was built correctly
[E] FAILED | Runtime: 108.198s | Command: D:\ProgramData\anaconda3\envs\bridgedepth\Scripts\polygraphy run E:\AI\Code\WorkCode\RT-Stereo\onnx\bridgedepth_eth3d_opset17_320_320_dynamic.onnx --trt --trt-min-shapes left_image:[1,3,32,320] right_image:[1,3,32,320] --trt-opt-shapes left_image:[1,3,320,320] right_image:[1,3,320,320] --trt-max-shapes left_image:[1,3,480,640] right_image:[1,3,480,640]

我使用netron定位/bridge_model/depth_head/output_conv1/Conv，找到了这一行代码
self.depth_feature_down = nn.Conv2d(depth_feature_dim, dim, kernel_size=4, stride=4, padding=0)
我不知道应该如何解决

Environment

TensorRT Version:10.3 and 10.13

NVIDIA GPU:RTX 2000 Ada

NVIDIA Driver Version:

CUDA Version:11.8

CUDNN Version:

Operating System:

Python Version (if applicable):3.10.18

Tensorflow Version (if applicable):

PyTorch Version (if applicable):2.7.0+cu118

Baremetal or Container (if so, version):

Relevant Files

Model link:

Steps To Reproduce

Commands or scripts:

Have you tried the latest release?:

Attach the captured .json and .bin files from TensorRT's API Capture tool if you're on an x86_64 Unix system

Can this model run on other frameworks? For example run ONNX model with ONNXRuntime (polygraphy run <model.onnx> --onnxrt):

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start with the attached bridgedepth.py and the reported trtexec failure around CaskConvolution and Cask convolution isConsistent. Reproduce the ONNX model with the Polygraphy ONNX Runtime and TensorRT commands at the reported input shapes, then compare the failing and passing cases. Done means identifying a reproducible TensorRT defect or a clear model/configuration cause and documenting the required correction.

Written by the indexing model from the issue text.

Assessment

Tech stack
cpp, python
Domain
backend, machine-learning
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Stale
Clarity
Needs clarification
Newbie friendliness
32/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.