OTB 7.1.0 has been released a few weeks ago. The binary packages are available here :
- OTB-7.1.0-Darwin64.run (Mac OS)
- OTB-7.1.0-Linux64.run (Linux)
- OTB-7.1.0-Win32.zip (Windows 32 bits)
- OTB-7.1.0-Win64.zip (Windows 64 bits)
It also possible to checkout the branch with git :
git clone https://gitlab.orfeo-toolbox.org/orfeotoolbox/otb.git OTB -b release-7.1
The full changelog can be accessed here. Let’s go through some of the most noticeable changes :
A conda package is now available for Orfeo Toolbox on Linux. It can be found on anaconda cloud and can be installed with the command
conda install -c orfeotoolbox otb
Please note that this is the first release of this package and it might not be completely stable yet. We welcome your feedbacks on OTB’s Gitlab.
OTB 7.0.0 brought compatibility with gdal 3, but version 2 was still used in the binary packages. In OTB 7.1.0 GDAL 3.0.2 and PROJ 6.2.1 are now used in the SuperBuild and in the binary packages.
One implication is that the “PROJ_LIB” environment must now be set before using OTB. It should point to the proj data dir. In the binary packages it is located at “Path/To/OTB/share/proj”. Note that the “otbenv.profile” script on UNIX and “otbenv.bat” on Windows take care of setting this environment variable.
However the QGIS plugin does not set this variable and therefore OTB 7.1.0 cannot be used in current versions of QGIS. A fix has been pushed on QGIS and will be available in QGIS 3.14.
OTB is able to write its output block by block, if the file formats of the output and the inputs allow it (for example geotiff can be read block by block, png cannot). This is useful when working on large images because the whole data doesn’t have to be loaded in memory.
Until this release however, applications with several output images were processing them one by one. For example in “DimensionalityReduction” it is possible to output the reduced image (“out”) and the image re-projected in the original space after the dimensionality reduction (“outinv”), but these two output are processed separately : the reduced image is written first, and then the reprojection is computed, without using the first output. With the multi-writer both outputs are processed at the same time, one block of the reprojected image is written every time a block of the reduced image is, saving a lot of computations !
Note that this functionality can be deactivated with extended filenames.
NL means smoothing
A new application, FastNLMeans, has been added to the toolbox. It is an implementation of the NL-means algorithm. There are now three applications to perform image Smoothing in OTB
- FastNLMeans : NL-means
- Smoothing : mean filter, gaussian filter and anistropic diffusion
- MeanShiftSmoothing : Smoothing step of the mean shift algorithm.
Thank to everyone who contributed to this release !
The OTB team.