Documentation on compiling with newer CUDA versions

[alexsikomas]

Summary

Adds a new documentation file BUILDING-CUDA-LINUX.md detailing how to compile and install ONNXRuntime for newer versions of CUDA, and then how to use this for obs-backgroundremoval.

Motivation

I couldn't find any direct guides on compiling ONNXRuntime and getting it to work with obs-backgroundremoval so I had to go through a lot of headache to figure it out. I wanted to share what I've learned, feel free to expand on it.

Verification

I successfully compiled ONNXRuntime and subsequently obs-backgroundremoval using these steps on:

• OS: Arch Linux 6.17.7-arch1-2
• CUDA: 13.0
• GPU: RTX 3060

Issues

It can take a while to load a model onto the GPU, this is only surprising because the same does not occur with CPU inference. Not sure if this is an issue or the expected behaviour.

RMBG and RobustVideoMatting both give errors, other models work as expected though.

This is the error RobustVideoMatting gives:

2025-11-17 14:18:11.562711140 [E:onnxruntime:background-removal-inference, tensorrt_execution_provider.h:90 log] [2025-11-17 14:18:11   ERROR] [graphShapeAnalyzer.cpp::checkCalculationStatusSanity::2126] Error Code 2: Internal Error (Assertion !isInFlight(p.second.symbolicRep) failed.  In checkCalculationStatusSanity at /_src/optimizer/shapeof/graphShapeAnalyzer.cpp:2126)
2025-11-17 14:18:11.562769910 [E:onnxruntime:background-removal-inference, tensorrt_execution_provider.h:90 log] [2025-11-17 14:18:11   ERROR] ModelImporter.cpp:135: While parsing node number 2 [Resize -> "389"]:
2025-11-17 14:18:11.564561504 [E:onnxruntime:background-removal-inference, tensorrt_execution_provider.h:90 log] [2025-11-17 14:18:11   ERROR] ModelImporter.cpp:138: --- Begin node ---
input: "src"
input: "386"
input: "388"
output: "389"
name: "Resize_3"
op_type: "Resize"
attribute {
  name: "extrapolation_value"
  f: 0
  type: FLOAT
}
attribute {
  name: "cubic_coeff_a"
  f: -0.75
  type: FLOAT
}
attribute {
  name: "nearest_mode"
  s: "floor"
  type: STRING
}
attribute {
  name: "mode"
  s: "linear"
  type: STRING
}
attribute {
  name: "coordinate_transformation_mode"
  s: "pytorch_half_pixel"
  type: STRING
}
attribute {
  name: "exclude_outside"
  i: 0
  type: INT
}

2025-11-17 14:18:11.564591405 [E:onnxruntime:background-removal-inference, tensorrt_execution_provider.h:90 log] [2025-11-17 14:18:11   ERROR] ModelImporter.cpp:139: --- End node ---
2025-11-17 14:18:11.564602365 [E:onnxruntime:background-removal-inference, tensorrt_execution_provider.h:90 log] [2025-11-17 14:18:11   ERROR] ModelImporter.cpp:141: ERROR: ModelImporter.cpp:368 In function parseNode:
[6] Invalid Node - Resize_3
[graphShapeAnalyzer.cpp::checkCalculationStatusSanity::2126] Error Code 2: Internal Error (Assertion !isInFlight(p.second.symbolicRep) failed.  In checkCalculationStatusSanity at /_src/optimizer/shapeof/graphShapeAnalyzer.cpp:2126)
2025-11-17 14:18:11.582141284 [E:onnxruntime:, inference_session.cc:2521 operator()] Exception during initialization: /home/alex/Documents/Applications/libs/onnxruntime/onnxruntime/core/providers/tensorrt/tensorrt_execution_provider.cc:2318 SubGraphCollection_t onnxruntime::TensorrtExecutionProvider::GetSupportedList(SubGraphCollection_t, int, int, const onnxruntime::GraphViewer&, bool*) const [ONNXRuntimeError] : 1 : FAIL : TensorRT input: 389 has no shape specified. Please run shape inference on the onnx model first. Details can be found in https://onnxruntime.ai/docs/execution-providers/TensorRT-ExecutionProvider.html#shape-inference-for-tensorrt-subgraphs

error: [obs-backgroundremoval] Exception during initialization: /home/alex/Documents/Applications/libs/onnxruntime/onnxruntime/core/providers/tensorrt/tensorrt_execution_provider.cc:2318 SubGraphCollection_t onnxruntime::TensorrtExecutionProvider::GetSupportedList(SubGraphCollection_t, int, int, const onnxruntime::GraphViewer&, bool*) const [ONNXRuntimeError] : 1 : FAIL : TensorRT input: 389 has no shape specified. Please run shape inference on the onnx model first. Details can be found in https://onnxruntime.ai/docs/execution-providers/TensorRT-ExecutionProvider.html#shape-inference-for-tensorrt-subgraphs

[umireon]

Thanks for the wonderful documentation!