Skip to content

Previous median_filter functionality and current float [-1,1] requirement #1616

@dshean

Description

@dshean

I just upgraded to 0.11.3 and I'm kind of surprised by the changes in the median filter functionality. I used the older skimage.filter.median_filter rather than the scipy.ndimage.filters.median_filter because I preferred the way it handled missing values. Scipy propagates data gaps (http://docs.astropy.org/en/v0.3/convolution/index.html), while the old skimage median filter would "fill" small gaps. It looks like the old skimage median_filter functionality is no longer available?

My primary inputs are large, single-band float32 or float64 numpy arrays (that often have data gaps). I initially avoided upgrading skimage because of the lack of support for float arrays. It looks like this has received some recent attention (#1419, #1529, #1422). I tried to update my tools with the new function, but then received the ValueError: Images of type float must be between -1 and 1.

It seems that requiring the user to scale their input float array to [-1,1] will be cumbersome. Hopefully this will happen behind the scenes in future releases.

Metadata

Metadata

Assignees

No one assigned

    Projects

    No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions