GstOnnxBaseInference
GObject ╰──GInitiallyUnowned ╰──GstObject ╰──GstElement ╰──GstBaseTransform ╰──GstOnnxBaseInference
Since : 1.30
Properties
device
“device” gchararray
Device identifier to use for inference, interpreted per execution provider. For CUDA, MIGraphX and HIP this is the integer device index (e.g. "0"). For Core ML this is the set of compute units the model can run on: "CPUOnly", "CPUAndGPU", "CPUAndNeuralEngine" or "ALL". When NULL (the default) the execution provider default is used.
Flags : Read / Write
Default value : NULL
Since : 1.30
execution-provider
“execution-provider” Onnx-execution-provider *
ONNX execution provider
Flags : Read / Write
Default value : cpu (0)
Since : 1.24
model-cache-dir
“model-cache-dir” gchararray
Directory used by the execution provider to cache compiled models. Currently supported by the MIGraphX, Vitis AI, HIP and Core ML execution providers.
Flags : Read / Write
Default value : NULL
Since : 1.30
model-file
“model-file” gchararray
Path to the ONNX model file.
A modelinfo file describing the model's input and output tensors is
required alongside it, and is looked up as <model-file>.modelinfo
first, then as <model-file without extension>.modelinfo. For
/path/model.onnx that is /path/model.onnx.modelinfo, then
/path/model.modelinfo.
Use the modelinfo-generator.py tool to generate one.
Flags : Read / Write
Default value : NULL
Since : 1.24
optimization-level
“optimization-level” Onnx-optimization-level *
ONNX optimization level
Flags : Read / Write
Default value : enable-extended (2)
Since : 1.24
vitisai-config-file
“vitisai-config-file” gchararray
Vitis AI execution provider configuration file. This is passed to the
provider as its config_file option when execution-provider is "vitisai".
Flags : Read / Write
Default value : NULL
Since : 1.30
Named constants
Onnx-execution-provider
Members
cpu (0) – CPU execution provider
cuda (1) – CUDA execution provider
vsi (2) – VeriSilicon NPU execution provider
migraphx (3) – AMD MIGraphX execution provider
hip (4) – AMD HIP execution provider (compiled out, requires ONNX Runtime >= 1.22)
dml (5) – Microsoft DirectML execution provider (compiled out, will error)
vitisai (6) – AMD Vitis AI execution provider
coreml (7) – Apple Core ML execution provider (compiled out, requires macOS and ONNX Runtime >= 1.21)
Onnx-optimization-level
Members
disable-all (0) – Disable all optimization
enable-basic (1) – Enable basic optimizations (redundant node removals)
enable-extended (2) – Enable extended optimizations (redundant node removals + node fusions)
enable-all (3) – Enable all possible optimizations
The results of the search are