onnxruntime_go

[go-dockly]

Hey I tried to load your onnx model into
https://github.com/yalue/onnxruntime_go
it is a much simpler onnx wrapper and loading networks is usually straight forward but with the silero vad onnx model something must be off in the way the header is encoded it complains that the shape info is not set:

	onnx.SetSharedLibraryPath(onnxLibPath)
	if err = onnx.InitializeEnvironment(); err != nil {
		return nil, fmt.Errorf("failed to init onnx lib: %w", err)
	}
	inputs, outputs, err := onnx.GetInputOutputInfo(netPath)
	if err != nil {
		return nil, fmt.Errorf("failed to get info for %s: %w", netPath, err)
	}
	for _, i := range inputs {
		fmt.Printf("	%s:%v ->", i.Name, i.Dimensions)
	}
	for _, o := range outputs {
		fmt.Printf("	%s\n", o.Name)
	}

        panic: runtime error: index out of range [0] with length 0
        
        goroutine 36 [running]:
        github.com/yalue/onnxruntime_go.getShapeFromInfo.func2(0x12c2394f0?, {0x1048b7000?, 0x140001bdb28?, 0x104445cd4?}, 0x12c210210?)
                /Users/enigma/go/pkg/mod/github.com/yalue/[email protected]/onnxruntime_go.go:1754 +0x88
        github.com/yalue/onnxruntime_go.getShapeFromInfo(0x12c2394f0)`

Would it be possible to make the network with standard way of encoding model meta data? Unfortunately, I don't know enough of python myself :(

[go-dockly]

https://github.com/onnx/models/blob/main/validated/vision/object_detection_segmentation/yolov4/dependencies/Conversion.ipynb

Found this example how they encode the onnx model from tflite for yolov4.