DenoiSeg requires keras 2.2.5 (as per README instructions, but also during pip install denoiseg). This version of keras contains a workaround for a specific h5py version, where dataset attributes are returned as bytes instead of str (see h5py/h5py#379).
Newer versions of h5py have this bug fixed, which breaks loading of model weights with keras 2.2.5 (attempts to decode a str).
Just conda installing keras 2.2.5 and h5py results in an incompatible combination of those two. I tried to conda install an older version of h5py (2.2.1) but gave up after one hour of conda sitting at "Solving environment"...
Could you maybe provide a conda yaml of an environment that is known to work?
DenoiSeg requires
keras2.2.5 (as per README instructions, but also duringpip install denoiseg). This version ofkerascontains a workaround for a specifich5pyversion, where dataset attributes are returned asbytesinstead ofstr(see h5py/h5py#379).Newer versions of
h5pyhave this bug fixed, which breaks loading of model weights withkeras2.2.5 (attempts todecodeastr).Just conda installing
keras2.2.5 andh5pyresults in an incompatible combination of those two. I tried to conda install an older version ofh5py(2.2.1) but gave up after one hour of conda sitting at "Solving environment"...Could you maybe provide a conda yaml of an environment that is known to work?