Tags: HumanBrainProject/neuroglancer-scripts
Tags
Version 1.2.0
New features
- Add support for the sharded precomputed format. Many thanks to Xiao
Gui for implementing this significant new feature in PR #35.
Other improvements
- Performance of the HTTP accessor was greatly improved by reusing a
HTTP session. Thanks to Xiao Gui for implementing this in PR #33.
Python version support
- Add support and tests for Python 3.12.
- Drop support for Python 3.5.
Bug fixes
- Fix incompatibilities with NumPy 2.0.0, which notably caused
incorrect data to be encoded when using the compressed_segmentation
encoding.
Version 1.1.0
New features
- Add a --compresslevel option for controlling the GZip compression
level. Many thanks to Bradley Lowekamp for implementing this feature
in PR #24 / PR #26.
- Add the --type and --encoding options to the all-in-one
volume-to-precomputed-pyramid script. Many thanks to Xiao Gui for
implementing this feature in PR #29.
Python version support
- Add support and tests for Python 3.11.
Bug fixes
- Fix volume reading with recent Nibabel >= 5.0.0, due to the expired
deprecation of get_data. Many thanks to Bradley Lowekamp for
reporting this bug, and to Xiao Gui for implementing the fix in PR
#28.
Version 1.0.0
New features
- Support for RGB NIfTI. Many thanks to Xiao Gui for implementing this
feature in PR #15.
- Support for LZW-compressed TIFF files. Many thanks to Xiao Gui for
implementing this feature in PR #14.
Python version support
- Add support and tests for Python 3.8, 3.9, and 3.10.
- Drop support for Python 3.4
Bug fixes
- Fix mesh conversion with recent versions of Nibabel (commit
c03bb5c).
- Bump the minimum version of tqdm to prevent a blocking crash (see
tqdm bug#613).
Version 0.3.0 released on PyPI
Bug fixes
- Fix the swarm of warning messages that appear when downscaling
images of integer type by the averaging method.
Other improvements
- The default downscaling method is now chosen automatically based on
the type of dataset: image uses local averaging, segmentation
uses striding.
- The command-line interface of mesh-to-precomputed was changed to
work with Accessors, instead of being restricted to files. The
command also now reads and writes the info file to make sure that
Neuroglancer knows where to find the mesh fragments.
- Introduced a new command link-mesh-fragments to create the small
JSON files that are required to link segmentation labels to
mesh fragments.
Version 0.2.0 of the PyPI package
Release Notes
=============
- Use correct rounding when downscaling by local average ([commit
8f77b48](8f77b48)).
- Fixed passing of options between functions ([issue
\#7](#7),
[commit
67430f1](67430f1),
[commit
f0e1e79](f0e1e79)).
Thanks to Ben Falk for reporting and fixing the issue.
- Fixed the conversion of stacks of multi-channel images ([commit
ff8d4dc](ff8d4dc)).
- Fixed a crash when flippping mesh triangles for negative-determinant
transformations ([commit
9754591](9754591),
[issue
\#5](#5)).
- Fixed a bug where the chunk size of extremely anisotropic volumes
was set to zero ([commit
92264c9](92264c9)).
- Fixed loading of `uint64` TIFF files ([issue
\#8](#8)).
Thanks to Ben Falk for reporting and fixing the issue.
- Introduced a new command `volume-to-precomputed-pyramid` for
all-in-one conversion of volume in simple cases.
- Introduced a new command `convert-chunks` which can be used to
convert existing data in Neuroglancer prec-computed format to a
different encoding.
- `slice-to-precomputed` now uses proper rounding and clamping for
performing data-type conversion.
- Improved the conversion of meshes to Neuroglancer-compatible VTK
format, and got rid of the dependency on
[PyVTK](https://github.com/pearu/pyvtk).
- Improved the performance of common use-cases by loading the full
volume in memory by default. The old behaviour can be restored with
`--mmap`.