FAQs:
Q: What is AlignEM-SWiFT?
A: AlignEM-SWiFt is a software tool specialized for registering electron micrographs. It is able to generate scale image hierarchies, compute affine transforms, and generate aligned images using multi-image rendering.
Q: Can AlignEM-SWiFT be used to register/"align" images other than electron micrographs?
A: Sure, however the Signal Whitening Image Matching (SWIM) was designed specifically with EM in mind. Such images are typically large in size and greyscale. AlignEM-SWIFT provides functionality for downscaling and the ability to pass alignment results (affines)
from lower scale levels to higher ones.
Q: What format is the output of AlignEM-SWiFT?
A: Scaled image series and generated alignments are always stored in two formats: TIFF and Zarr.Zarr is an open-source format for the storage of chunked, compressed, N-dimensional arrays with an interface similar to NumPy. It has a Nature Methods paper:
https://www.nature.com/articles/s41592-021-01326-w
For more information about Zarr: Help > Zarr Help
Q: What is meant by "scales"?
A: In AlignEM-SWiFT a "scale" simply means a downsampled (of decreased resolution) copy of an image series.
Q: Why should data be scaled at all? Is it okay to align the full resolution series with brute force?
A: You could, but EM images tend to run large. A more efficient workflow is to:
1) generate a hierarchy of downsampled images from the full resolution images
2) align the lowest resolution images first
3) pass the computed affines to the scale of next-highest resolution, and repeat until the full resolution images are in alignment. In these FAQs this is referred to as "climbing the scale hierarchy""
Q: Is it possible to generate a new scale pyramid without starting a new project?
A: Yes. To rescale your dataset use the menu option Actions > Rescale... Caution (!) - Rescaling is effectually like starting over but using the same images imported previously.
Q: Why do SNR values not necessarily increase as we "climb the scale hierarchy"?
A: SNR values returned by SWIM are a relativistic readout of alignment quality which depends on image resolution. It is most useful when comparing the performance of the SWIM image matching algorithm between stationary/moving image pairs at the same scale.
Q: Why are the selected manual correlation regions not mutually independent? In other words, why does moving or removing an argument to SWIM affect the signal-to-noise ratio and resulting correlation signals of the other selected SWIM regions?
A: This is a consequence of the fact that SWIM is an iterative algorithm as it is implemented for the Default Grid, Custom Grid, and Manual Hint alignment methods
Q: What is Neuroglancer?
A: Neuroglancer is an open-source WebGL and typescript-based web application for displaying volumetric data. AlignEM-SWiFT uses a Chromium-based API called QtWebEngine together with the Neuroglancer Python API to render large volumetric data efficiently and conveniently within the application window.
Q: What makes AlignEM-SWiFT fast?
A: A few reasons:
1) Multi-core processing, time-intensive processes are executed in parallel.
2) Data scaling, SWIM alignment, and affine processing functions are all implemented in highly efficient C code written by computer scientist Arthur Wetzel.
3) Fast Fourier Transform is a fast algorithm.
Q: How many CPUs or "cores" will AlignEM-SWiFT use?
A: By default, as many cores as the system has available (greedy).
Q: What file types are supported?
A: Currently, only images formatted as TIFF are supported.
Q: Where can I learn more about the principles of Signal Whitening Fourier Transform Image Matching?
A: https://mmbios.pitt.edu/images/ScientificMeetings/MMBIOS-Aug2014.pdf