`UndefVarError: stride=(2, 2) not defined` when loading ResNet

[natema]

I downloaded resnet18 from here, and run, in the onnx file's folder

ONNX.load_model("resnet18-v1-7.onnx")
weights = ONNX.load_weights("weights.bson") ;
model = include("model.jl")

which results in

julia> model = include("model.jl")
ERROR: LoadError: UndefVarError: stride=(2, 2) not defined
Stacktrace:
 [1] top-level scope at /mypath🙂/models/model.jl:6
 [2] include(::String) at ./client.jl:439
 [3] top-level scope at REPL[44]:1
in expression starting at /mypath🙂/models/model.jl:4

Any idea?

[natema]

I made some (apparent) progress by looking at this issue.

In the model.jl file produced by this package, I removed the enclosing var"..." from all statements of the type
var"stride=(2, 2)", var"pad=(3, 3, 3, 3)" and var"dilation=(1, 1)".
(I also tried to comment the function Mul and Add, which appear irrelevant.)
Now when I try to feed to the model a zeros(224,224,3,1) tensor, it complains that

ERROR: ArgumentError: At least one of `length` or `step` must be specified
Stacktrace:
 [1] _range2(::Int64, ::Nothing, ::Int64, ::Nothing) at ./range.jl:94
 [2] range(::Int64, ::Int64; length::Nothing, step::Nothing) at ./range.jl:91
 [3] range(::Int64, ::Int64) at ./range.jl:91
 [4] (::var"#25#26")(::Array{Float64,4}) at /😎/models/model.jl:59
 [5] top-level scope at REPL[31]:1

[Paethon]

Same problem when loading ResNet50 with error UndefVarError: stride=(1, 1) not defined