Thoughts.txt

TODO:

Update command line syntax to be more flexible, like ImageMagick
 e.g. convertonnxmodel convert (dataType:float32 dimensions:[256,256,3,1] strides:ONNX dimensionSemantics:NCHW input:foo.dat) cast:float32 scale:255 normalize:true output:bar.dat
      convertonnxmodel convert dataType:float32 dimensions:[256,256,3,1] strides:ONNX foo.dat cast:float32 scale:255 normalize:true bar.dat

TODO: Print more detailed input and output information.

convertonnxmodel convert input.onnxtensor cast:float32 normalize:true scale:255 layout:WHC strides:increasing output.bin

input:{
    file:"input.onnxtensor"
    name:"MulWeights_42"
    datatype:uint8
    dimensions:[3,240,320]
    strides:[307200,320,1] // decreasing ONNX
    layout:CHW
}
operations:{
    cast:float32
    normalize:true
    scale:255
}
output:{
    file:"output.bin"
    datatype:float32
    dimensions:[320,240,3]
    strides:[1,3,640]
    layout:WHC
}

convertonnxmodel convert input.onnxtensor output.png

input:{
    file:"input.onnxtensor"
    datatype:uint8
    dimensions:[3,240,320]
    strides:[307200,320,1] // decreasing
    layout:CHW
}
operations:{
    cast:uint8
}
output:{
    file:"output.png"
    datatype:uint8
    dimensions:[3,320,240]
    bytestrides:[1,3,640] // increasing
    layout:CWH
}

convertonnxmodel convert input.csv size:3,320,240 output.png

input:{
    file:"input.csv"
    datatype:uint8
    dimensions:[230400]
    strides:[1] // increasing
    layout:_
}
operations:{
    cast:uint8 // implicit
    reshape:{size:[3,320,240] strides:increasing} // implicit
}
output:{
    file:"output.png"
    datatype:uint8
    dimensions:[3,320,240]
    strides:[1,3,640] // increasing
    layout:CWH
}

See Protobuf CMake here: https://www.reddit.com/r/cpp/comments/17v0tui/exploring_cmake_and_protobuf_my_tiny_project_for/