Simple reader for (non-fancy) neural networks in .onnx format for Julia.

The currently supported operations are: Add, Sub, MatMul, Gemm, Relu. (Flatten is not supported, but is ignored as long as only one dimension is larger than one.)

The code is based on the python implementation in NNENUM and just uses PyCall.jl to interface to python.

Installation

Type ] in the Julia REPL to activate the package manager. The OnnxReader can then be installed via

(v1.0) pkg> add https://github.com/phK3/OnnxReader.jl

(Make sure onnx and numpy are available in the Python environment used by PyCall.jl)

Example

using OnnxReader

ws, bs = load_network("./example_nns/ACASXU_run2a_1_1_batch_2000.onnx")
@show size(ws)
@show size(bs)

Should return the weights and biases of the respective network.