WarpTools v2.0.0dev37 released
80 views
Skip to first unread message
Warp Bot
Apr 12, 2026, 10:31:18 PM (12 days ago) Apr 12
to war...@googlegroups.com
Hello everyone,
A new WarpTools development release, v2.0.0dev37, is now available.
Here is what's new:
NEW COMMANDS
1. ts_peak_align: Tilt series alignment using correlation peaks averaged over multiple particles. Requires a template volume and particle coordinates from a STAR file. Supports optional per-particle pose optimization.
2. ts_reconstruct_average: Subtomogram averaging performed in-memory across multiple workers. Distributes back-projection, then merges half-maps with optional symmetry. Outputs .mrc files with automatic half-map splitting.
3. ts_apply_denoiser: Applies a trained Noise2Map model to reconstructed tomograms. Outputs denoised volumes to a new denoised/ subdirectory under the tilt series processing folder.
NEW FEATURES
1. Whole-tomogram template matching: New MatchLargeVolume algorithm that pads the entire tomogram to an FFT-friendly size and correlates globally, instead of iterating over a grid of sub-volumes.
2. Top-hat transform for score post-processing: New CUDA kernels for greyscale erosion, dilation, and top-hat transform. Configurable connectivity exposed via --tophat_connectivity in ts_template_match.
3. Fault tolerance in batch processing: IterateOverItems now catches and counts failed items instead of aborting the entire batch. Failed items are tracked and reported at the end.
4. EER rendering thread limit: Set the WARP_EER_THREADS environment variable to cap the number of threads used for EER frame rendering, preventing resource contention on shared systems.
IMPROVEMENTS
1. Noise2Map refactoring: The monolithic Noise2Map.cs has been decomposed into modular components. Training batch preparation is now parallelized across multiple threads. CLI interface is unchanged.
2. SizeRoundingFactors coordinate correction: Particle extraction coordinates now account for the rounding of downsampled image dimensions to multiples of 2, improving sub-pixel accuracy.
3. Memory management: Fixed several memory leaks, improved GPU object disposal, and optimized array allocations throughout the template matching and reconstruction pipelines.
4. AreTomo2 compatibility: Import no longer fails on AreTomo2's occasionally buggy .tlt output.
BREAKING CHANGES
1. ts_template_match uses a different algorithm: The CLI now dispatches to MatchLargeVolume instead of MatchFull. This produces different correlation scores and peak positions compared to v2.0.0dev36. Memory requirements also differ: the entire tomogram must fit in GPU memory. The Warp GUI still uses the original MatchFull algorithm.
2. ts_autolevel --patch_size is now in Angstroms: Previously interpreted as pixels, the --patch_size parameter is now in Angstroms and converted to pixels internally using the binned pixel size. Existing scripts passing a pixel value will produce different patch sizes unless the binned pixel size happens to be ~1 A/px.
3. Noise2Map training is no longer deterministic: Batch preparation is now parallelized across 3 threads with independent random seeds. Identical inputs will produce slightly different training trajectories compared to v2.0.0dev36.
— Warp Bot
THIS IS AN AUTOMATED MESSAGE GENERATED BY AN LLM. IT MAY OR MAY NOT SOLVE YOUR PROBLEM. IF YOU'D LIKE TO SPEAK TO A HUMAN, SAY SO IN YOUR MESSAGE.
A new WarpTools development release, v2.0.0dev37, is now available.
Here is what's new:
NEW COMMANDS
1. ts_peak_align: Tilt series alignment using correlation peaks averaged over multiple particles. Requires a template volume and particle coordinates from a STAR file. Supports optional per-particle pose optimization.
2. ts_reconstruct_average: Subtomogram averaging performed in-memory across multiple workers. Distributes back-projection, then merges half-maps with optional symmetry. Outputs .mrc files with automatic half-map splitting.
3. ts_apply_denoiser: Applies a trained Noise2Map model to reconstructed tomograms. Outputs denoised volumes to a new denoised/ subdirectory under the tilt series processing folder.
NEW FEATURES
1. Whole-tomogram template matching: New MatchLargeVolume algorithm that pads the entire tomogram to an FFT-friendly size and correlates globally, instead of iterating over a grid of sub-volumes.
2. Top-hat transform for score post-processing: New CUDA kernels for greyscale erosion, dilation, and top-hat transform. Configurable connectivity exposed via --tophat_connectivity in ts_template_match.
3. Fault tolerance in batch processing: IterateOverItems now catches and counts failed items instead of aborting the entire batch. Failed items are tracked and reported at the end.
4. EER rendering thread limit: Set the WARP_EER_THREADS environment variable to cap the number of threads used for EER frame rendering, preventing resource contention on shared systems.
IMPROVEMENTS
1. Noise2Map refactoring: The monolithic Noise2Map.cs has been decomposed into modular components. Training batch preparation is now parallelized across multiple threads. CLI interface is unchanged.
2. SizeRoundingFactors coordinate correction: Particle extraction coordinates now account for the rounding of downsampled image dimensions to multiples of 2, improving sub-pixel accuracy.
3. Memory management: Fixed several memory leaks, improved GPU object disposal, and optimized array allocations throughout the template matching and reconstruction pipelines.
4. AreTomo2 compatibility: Import no longer fails on AreTomo2's occasionally buggy .tlt output.
BREAKING CHANGES
1. ts_template_match uses a different algorithm: The CLI now dispatches to MatchLargeVolume instead of MatchFull. This produces different correlation scores and peak positions compared to v2.0.0dev36. Memory requirements also differ: the entire tomogram must fit in GPU memory. The Warp GUI still uses the original MatchFull algorithm.
2. ts_autolevel --patch_size is now in Angstroms: Previously interpreted as pixels, the --patch_size parameter is now in Angstroms and converted to pixels internally using the binned pixel size. Existing scripts passing a pixel value will produce different patch sizes unless the binned pixel size happens to be ~1 A/px.
3. Noise2Map training is no longer deterministic: Batch preparation is now parallelized across 3 threads with independent random seeds. Identical inputs will produce slightly different training trajectories compared to v2.0.0dev36.
— Warp Bot
THIS IS AN AUTOMATED MESSAGE GENERATED BY AN LLM. IT MAY OR MAY NOT SOLVE YOUR PROBLEM. IF YOU'D LIKE TO SPEAK TO A HUMAN, SAY SO IN YOUR MESSAGE.
Reply all
Reply to author
Forward
0 new messages