PyObject support
Currently in my tests, it is very slow to transfer big numpy arrays, like from a camera into a mojo Tensor, will this case have better support in the future?
3 Replies
I assume this is via CPython, yes if there will be ways to convert the structures directly in compiled mojo code.
I mean that why you do like
let np = Python.import_module("numpy")
let x = np.array([1, 2, 3])
Than x is PyObject, and if you want to iterate x fastly you would rather x to be Tensor, I am asking if there will be an intended way that will work fast to transfer PyObject, eve only numpy arrays into moji's Tensor
I'm sure something like this will happen in future, might be that parts of NumPy might support mojo directly.