Fail to make TensorRT model from onnx in Isaac ROS Image Segment Tutorial

[JimXu1989]

https://nvidia-isaac-ros.github.io/repositories_and_packages/isaac_ros_image_segmentation/isaac_ros_segformer/index.html#quickstart
I fellow the tutorial and the command "/usr/src/tensorrt/bin/trtexec --onnx=${ISAAC_ROS_WS}/isaac_ros_assets/models/peoplesemsegformer/1/model.onnx --saveEngine=${ISAAC_ROS_WS}/isaac_ros_assets/models/peoplesemsegformer/1/model.plan"

show the error:

(isaac-ros) jim@jim-ROG-Strix-G834JZ-G834JZ:~/workspaces/isaac_ros-dev/isaac_ros_assets/models/peoplesemsegformer/1$ cd ${ISAAC_ROS_WS}/isaac_ros_assets/models/peoplesemsegformer/1

rm -f model_int64.plan

/usr/src/tensorrt/bin/trtexec
--onnx=model.onnx
--saveEngine=model_int64.plan
--shapes=input:1x3x512x512
--outputIOFormats=int64:chw
2>&1 | tee /tmp/trtexec_int64.log

ls -lh model_int64.plan
&&&& RUNNING TensorRT.trtexec [TensorRT v101601] [b11] # /usr/src/tensorrt/bin/trtexec --onnx=model.onnx --saveEngine=model_int64.plan --shapes=input:1x3x512x512 --outputIOFormats=int64:chw
[05/04/2026-16:47:41] [I] TF32 is enabled by default. Add --noTF32 flag to further improve accuracy with some performance cost.
[05/04/2026-16:47:41] [I] === Model Options ===
[05/04/2026-16:47:41] [I] Format: ONNX
[05/04/2026-16:47:41] [I] Model: model.onnx
[05/04/2026-16:47:41] [I] Output:
[05/04/2026-16:47:41] [I] === Build Options ===
[05/04/2026-16:47:41] [I] Memory Pools: workspace: default, dlaSRAM: default, dlaLocalDRAM: default, dlaGlobalDRAM: default, tacticSharedMem: default
[05/04/2026-16:47:41] [I] avgTiming: 8
[05/04/2026-16:47:41] [I] Precision: FP32
[05/04/2026-16:47:41] [I] LayerPrecisions:
[05/04/2026-16:47:41] [I] Layer Device Types:
[05/04/2026-16:47:41] [I] Decomposable Attentions:
[05/04/2026-16:47:41] [I] Calibration:
[05/04/2026-16:47:41] [I] Refit: Disabled
[05/04/2026-16:47:41] [I] Strip weights: Disabled
[05/04/2026-16:47:41] [I] Version Compatible: Disabled
[05/04/2026-16:47:41] [I] ONNX Plugin InstanceNorm: Disabled
[05/04/2026-16:47:41] [I] ONNX kENABLE_UINT8_AND_ASYMMETRIC_QUANTIZATION_DLA flag: Disabled
[05/04/2026-16:47:41] [I] ONNX kREPORT_CAPABILITY_DLA flag: Disabled
[05/04/2026-16:47:41] [I] ONNX kADJUST_FOR_DLA flag: Disabled
[05/04/2026-16:47:41] [I] ONNX kENABLE_PLUGIN_OVERRIDE flag: Disabled
[05/04/2026-16:47:41] [I] TensorRT runtime: full
[05/04/2026-16:47:41] [I] Lean DLL Path:
[05/04/2026-16:47:41] [I] Tempfile Controls: { in_memory: allow, temporary: allow }
[05/04/2026-16:47:41] [I] Exclude Lean Runtime: Disabled
[05/04/2026-16:47:41] [I] Sparsity: Disabled
[05/04/2026-16:47:41] [I] Safe mode: Disabled
[05/04/2026-16:47:41] [I] Build DLA standalone loadable: Disabled
[05/04/2026-16:47:41] [I] Allow GPU fallback for DLA: Disabled
[05/04/2026-16:47:41] [I] DirectIO mode: Disabled
[05/04/2026-16:47:41] [I] Restricted mode: Disabled
[05/04/2026-16:47:41] [I] Skip inference: Disabled
[05/04/2026-16:47:41] [I] Save engine: model_int64.plan
[05/04/2026-16:47:41] [I] Load engine:
[05/04/2026-16:47:41] [I] Profiling verbosity: 0
[05/04/2026-16:47:41] [I] Tactic sources: Using default tactic sources
[05/04/2026-16:47:41] [I] timingCacheMode: local
[05/04/2026-16:47:41] [I] timingCacheFile:
[05/04/2026-16:47:41] [I] Enable Compilation Cache: Enabled
[05/04/2026-16:47:41] [I] Enable Monitor Memory: Disabled
[05/04/2026-16:47:41] [I] CPU Only Mode: Disabled
[05/04/2026-16:47:41] [I] errorOnTimingCacheMiss: Disabled
[05/04/2026-16:47:41] [I] Preview Features: Use default preview flags.
[05/04/2026-16:47:41] [I] MaxAuxStreams: -1
[05/04/2026-16:47:41] [I] BuilderOptimizationLevel: -1
[05/04/2026-16:47:41] [I] MaxTactics: -1
[05/04/2026-16:47:41] [I] Calibration Profile Index: 0
[05/04/2026-16:47:41] [I] Weight Streaming: Disabled
[05/04/2026-16:47:41] [I] Runtime Platform: Same As Build
[05/04/2026-16:47:41] [I] Debug Tensors:
[05/04/2026-16:47:41] [I] Distributive Independence: Disabled
[05/04/2026-16:47:41] [I] Mark Unfused Tensors As Debug Tensors: Disabled
[05/04/2026-16:47:41] [I] Input(s)s format: fp32:CHW
[05/04/2026-16:47:41] [I] Output(s): int64:chw
[05/04/2026-16:47:41] [I] Input build shape (profile 0): input=1x3x512x512+1x3x512x512+1x3x512x512
[05/04/2026-16:47:41] [I] Input calibration shapes: model
[05/04/2026-16:47:41] [I] === System Options ===
[05/04/2026-16:47:41] [I] Device: 0
[05/04/2026-16:47:41] [I] DLACore:
[05/04/2026-16:47:41] [I] Loaded static plugins:
[05/04/2026-16:47:41] [I] setPluginsToSerialize:
[05/04/2026-16:47:41] [I] dynamicPlugins:
[05/04/2026-16:47:41] [I] ignoreParsedPluginLibs: 0
[05/04/2026-16:47:41] [I]
[05/04/2026-16:47:41] [I] === Inference Options ===
[05/04/2026-16:47:41] [I] Batch: Explicit
[05/04/2026-16:47:41] [I] Input inference shape : input=1x3x512x512
[05/04/2026-16:47:41] [I] Iterations: 10
[05/04/2026-16:47:41] [I] Duration: 3s (+ 200ms warm up)
[05/04/2026-16:47:41] [I] Sleep time: 0ms
[05/04/2026-16:47:41] [I] Idle time: 0ms
[05/04/2026-16:47:41] [I] Inference Streams: 1
[05/04/2026-16:47:41] [I] ExposeDMA: Disabled
[05/04/2026-16:47:41] [I] Data transfers: Enabled
[05/04/2026-16:47:41] [I] Spin-wait: Disabled
[05/04/2026-16:47:41] [I] Multithreading: Disabled
[05/04/2026-16:47:41] [I] CUDA Graph: Disabled
[05/04/2026-16:47:41] [I] Separate profiling: Disabled
[05/04/2026-16:47:41] [I] Time Deserialize: Disabled
[05/04/2026-16:47:41] [I] Time Refit: Disabled
[05/04/2026-16:47:41] [I] NVTX verbosity: 0
[05/04/2026-16:47:41] [I] Persistent Cache Ratio: 0
[05/04/2026-16:47:41] [I] Optimization Profile Index: 0
[05/04/2026-16:47:41] [I] Weight Streaming Budget: 100.000000%
[05/04/2026-16:47:41] [I] Inputs:
[05/04/2026-16:47:41] [I] Debug Tensor Save Destinations:
[05/04/2026-16:47:41] [I] Dump All Debug Tensor in Formats:
[05/04/2026-16:47:41] [I] === Reporting Options ===
[05/04/2026-16:47:41] [I] Verbose: Disabled
[05/04/2026-16:47:41] [I] Averages: 10 inferences
[05/04/2026-16:47:41] [I] Percentiles: 90,95,99
[05/04/2026-16:47:41] [I] Dump refittable layers:Disabled
[05/04/2026-16:47:41] [I] Dump output: Disabled
[05/04/2026-16:47:41] [I] Profile: Disabled
[05/04/2026-16:47:41] [I] Export timing to JSON file:
[05/04/2026-16:47:41] [I] Export output to JSON file:
[05/04/2026-16:47:41] [I] Export profile to JSON file:
[05/04/2026-16:47:41] [I]
[05/04/2026-16:47:41] [I] === Device Information ===
[05/04/2026-16:47:41] [I] Available Devices:
[05/04/2026-16:47:41] [I] Device 0: "NVIDIA GeForce RTX 4080 Laptop GPU" UUID: GPU-7d2d621d-c5b1-2cd9-91df-d67292816df5
[05/04/2026-16:47:41] [I] Selected Device: NVIDIA GeForce RTX 4080 Laptop GPU
[05/04/2026-16:47:41] [I] Selected Device ID: 0
[05/04/2026-16:47:41] [I] Selected Device UUID: GPU-7d2d621d-c5b1-2cd9-91df-d67292816df5
[05/04/2026-16:47:41] [I] Compute Capability: 8.9
[05/04/2026-16:47:41] [I] SMs: 58
[05/04/2026-16:47:41] [I] Device Global Memory: 11874 MiB
[05/04/2026-16:47:41] [I] Shared Memory per SM: 100 KiB
[05/04/2026-16:47:41] [I] Memory Bus Width: 192 bits (ECC disabled)
[05/04/2026-16:47:41] [I] Application Compute Clock Rate: 2.28 GHz
[05/04/2026-16:47:41] [I] Application Memory Clock Rate: 9.001 GHz
[05/04/2026-16:47:41] [I]
[05/04/2026-16:47:41] [I] Note: The application clock rates do not reflect the actual clock rates that the GPU is currently running at.
[05/04/2026-16:47:41] [I]
[05/04/2026-16:47:41] [I] TensorRT version: 10.16.1
[05/04/2026-16:47:41] [I] Loading standard plugins
[05/04/2026-16:47:41] [I] [TRT] [MemUsageChange] Init CUDA: CPU +2, GPU +0, now: CPU 32, GPU 249 (MiB)
[05/04/2026-16:47:43] [I] [TRT] [MemUsageChange] Init builder kernel library: CPU +2775, GPU +446, now: CPU 3009, GPU 695 (MiB)
[05/04/2026-16:47:43] [I] Start parsing network model.
[05/04/2026-16:47:43] [I] [TRT] ----------------------------------------------------------------
[05/04/2026-16:47:43] [I] [TRT] Input filename: model.onnx
[05/04/2026-16:47:43] [I] [TRT] ONNX IR version: 0.0.8
[05/04/2026-16:47:43] [I] [TRT] Opset version: 16
[05/04/2026-16:47:43] [I] [TRT] Producer name: pytorch
[05/04/2026-16:47:43] [I] [TRT] Producer version: 1.14.0
[05/04/2026-16:47:43] [I] [TRT] Domain:
[05/04/2026-16:47:43] [I] [TRT] Model version: 0
[05/04/2026-16:47:43] [I] [TRT] Doc string:
[05/04/2026-16:47:43] [I] [TRT] ----------------------------------------------------------------
[05/04/2026-16:47:43] [W] [TRT] ModelImporter.cpp:804: Make sure output output has Int64 binding.
[05/04/2026-16:47:43] [I] Finished parsing network model. Parse time: 0.188357
[05/04/2026-16:47:43] [I] Set shape of input tensor input for optimization profile 0 to: MIN=1x3x512x512 OPT=1x3x512x512 MAX=1x3x512x512
[05/04/2026-16:47:43] [I] Created engine with size: 0 MiB
[05/04/2026-16:47:43] [I] Engine built in 6.722e-05 sec.
[05/04/2026-16:47:43] [E] Assertion failure: false && "Attempting to access an empty engine!"
-rw-rw-r-- 1 jim jim 0 May 4 16:47 model_int64.plan

[jaiveersinghNV]

Hi @JimXu1989 ,

Based on your logs, I see that you're using TensorRT 10.16:

[05/04/2026-16:47:41] [I] TensorRT version: 10.16.1

However, Isaac ROS 4.4 only supports TensorRT 10.13. Please see the instructions for initializing the Isaac ROS venv environment isolation mode:
https://nvidia-isaac-ros.github.io/getting_started/index.html#initialize-isaac-ros-cli:~:text=OpenCV%204.6.0-,TensorRT%2010.13.3.9,-If%20you%20have

Please make sure to install the correct version of TensorRT on your system, and then try again. You may also use the Docker environment isolation mode to run Isaac ROS within a separate container, so that your host system can continue to use TensorRT 10.16.

[JimXu1989]

Hi @JimXu1989 ,

Based on your logs, I see that you're using TensorRT 10.16:

[05/04/2026-16:47:41] [I] TensorRT version: 10.16.1

However, Isaac ROS 4.4 only supports TensorRT 10.13. Please see the instructions for initializing the Isaac ROS venv environment isolation mode: https://nvidia-isaac-ros.github.io/getting_started/index.html#initialize-isaac-ros-cli:~:text=OpenCV%204.6.0-,TensorRT%2010.13.3.9,-If%20you%20have

Please make sure to install the correct version of TensorRT on your system, and then try again. You may also use the Docker environment isolation mode to run Isaac ROS within a separate container, so that your host system can continue to use TensorRT 10.16.

The same error with tensorRT 10.13

[jaiveersinghNV]

Could you please provide updated logs for the 10.13 run?