imresize throw StackOverflowError with sz of array type about imagetransformations.jl HOT 3 CLOSED

👍 , thanks indeed @johnnychen94.

But we should use a keyword argument for ratio. Otherwise it has ambiguous intent for one-dimensional "images." (JuliaImages has a much firmer commitment to providing good support for images of arbitrary dimensionality than does Matlab.) Or we could use Percent but I think a keyword argument would be better.

from imagetransformations.jl.

Update:

The document of imresize doesn't say sz must be Tuple{Vararg{Integer}} type, so I just try the MATLAB way of using imresize, and it throws an unexpected StackOverflowError.

• use [128,128] as size, it's Array type.
• use ceil.(size(img) .* 0.5) as size, it is Tuple{Float64,Float64} Type

using Images, TestImages
lena = testimage("lena_gray_256")

imresize(lena, (128, 128)) # work as expected
imresize(lena, [128, 128]) # throw StackOverflowError
imresize(lena, (128.0, 128.0)) # throw StackOverflowError

With a temporary solution, just use

using Images, TestImages
lena = testimage("lena_gray_256")

imresize(lena, (128, 128)) # work as expected
imresize(lena, ceil.(Int, size(lena) .* 0.5)) # work as expected

Also,
can we add an imresize(img, ratio) API support to this function just like MATLAB? Since Images are always at least 2D, there will be no ambiguity. I think it might be fairly easy to implement.

imresize(original::AbstractArray, dim1, dimN...) = imresize(original, (dim1,dimN...)) 
imresize(original::AbstractArray, ratio) = imresize(original, round.(Int, size(original).* ratio) # add this line of code

from imagetransformations.jl.

Thanks for the thorough report. I agree the current situation is suboptimal. I also like the ratio idea

from imagetransformations.jl.