CHANGES
=======

v0.0.5
------

* Update documentation for sample()
* Allow non-square displacement fields
* Clean up identity mapping code
* Remove caching of identity mappings
* Use align\_corners=False from PyTorch 1.3.0 Will no longer work with earlier PyTorch versions
* Convert to using MIT License

v0.0.4.post1
------------

* Prevent producing NaN in inverse backward pass
* Allow accessing field type as torchfields.Field
* Ensure contiguous gradients in inversion backward pass
* Allow padding to be explicitly given in \_pad()
* Bump required pytorch version to 1.1.0
* Factor out voting, inversion, and util functions
* Update README.md
* Update README.md

v0.0.3
------

* Change ndim to ndimension() to support wider range of PyTorch versions

v0.0.2
------

* [Fix] affine\_field incorrect dimensions bug

v0.0.1
------

* [Fix] inverse not working on cpu
* Add setup.py, setup.cfg, README.md, .gitignore, and requirements.txt
* DisplacementField: minor bug fixes, refactors, and comments
* Safe division to avoid NaNs during backward pass
* Use winding number to test inclusion rather than bounding i,j
* Epsilon for comparison to zero
* Autopad functionality in left inverse
* More efficient left inverse using sparse tensors
* Mean finite vector function
* Inefficient implementation of left inverse
* Add DisplacementField class to abstract displacement field operations
* Initial commit
