Changes

Version 0.6.0

tomotwim_embed.py is now 1.6x faster and linearly scales across multiple GPUs
- Exploiting new the compile option of the latest pytorch 2.1 nightly build.
- Internally DistributedDataParallel is used instead of DataParallel

Speed up embedding using Masks
- The command tomotwin_embed.py tomogram now has an optional --mask argument to select the region of interest for embedding.
- The command tomotwin_tools.py embedding_mask now computes a by isonet inspired mask that hides some parts of the tomogram volume that are unlikely to contain proteins. If you use the generated mask with new --mask argument, the embedding step is up to 2 times faster. CAUTION: In TomoTwin 0.4, the embeddings_mask command calculated a label mask for the clustering workflow. This functionality now happens automatically during the calculation of the umap (tomotwin_tools.py umap).
- Thanks Caitie McCafferty and Ricardo Righetto for the feature request
More accurate cluster centers
- When selecting clusters in Napari during the clustering workflow, the Medoid is now calculated instead of the average of all cluster embeddings. This has the advantage that it is guaranteed to be on the embedding hypersphere and should be a better representation of the cluster center than the average.
- The coordinates of the found medoid for each cluster is written as a .coords file to disk.
Other
- Updated installation instructions for napari: Napari 0.4.17 -> Napari 0.4.18
- Added some information snippets for developers

Fix numba related issue by updating various dependencies (https://github.com/MPI-Dortmund/tomotwin-cryoet/issues/20):
- Python 3.9 -> Python 3.10
- Rapids 22.04 -> 23.04
- CUDA 11.6 -> CUDA 11.8
Updating pyStarDB from 0.3.2 -> 0.4.2

Official clustering workflow release. Please checkout the updated installation instructions and in depth tutorial.
Added important tools like tomotwin_tools.py umap and tomotwin_tools.py embeddings_mask
Added more unit tests

Scale heatmaps to the same size as the tomogram, to make them overlay easily. Additionally, make them optional (–write_heatmaps) as they require some space
Write Relion 3 STAR files as they are required for WARP (Thanks Tom Dendooven)
Reference refinement deactivated by default, as we noticed that it makes the results worse in some cases.