zebsdsolve

EBSD pattern-centre fitting and indexing on the GPU, in the browser or from Python.

Open the web app: it runs on your GPU inside the page (Chrome or Edge 137+). Scans are read in place, in pieces, so a 100 GB file is neither uploaded nor loaded into memory. Click Example for a small nickel scan, or Make toy scan to simulate one from a master pattern.

What it does

• Indexes by spherical cross-correlation over all orientations, refined on the detector: about 23 000 patterns per second on a desktop GPU, 67 000 in two sweeps. Dictionary indexing is there too.
• Fits the geometry: the pattern centre by a vote of a few patterns, then the detector's whole pose (position, rotation and, optionally, pixel size) together with the patterns' orientations.
• Prepares patterns: binning, neighbour averaging (NLPAR), static and dynamic background, CLAHE, band pass, mask.
• Handles several phases, and writes .ang and .ctf.
• Reads h5ebsd (kikuchipy, Bruker, EDAX, Oxford .h5oina), EDAX .up1 / .up2, Oxford .ebsp, NORDIF Pattern.dat. Master patterns: ebsdsim .npz and EMsoft .h5.

One engine (Zig and WGSL) runs natively through wgpu-native and as WebAssembly on WebGPU, so the app and the Python package do the same thing.

Python

pip install zebsdsolve
python -m zebsdsolve          # the same app, served locally

import zebsdsolve as ze

res = ze.index("scan.h5", ["Ni.npz"], fit="pose")    # vote, pose fit, then index
res.quaternions, res.ncc, res.phase                  # per map cell
res.geometry                                         # what the fits found

Every option is a keyword (src/settings.zig); ze.Solver drives the steps one at a time. Two tutorials are in examples/.

More

• How it works, and what was measured: each method, the numbers behind the defaults, and what is not verified.
• CONTRIBUTING.md: building, testing, conventions.

MIT (LICENSE). The wheels bundle wgpu-native (MIT / Apache-2.0); the example scan is kikuchipy's nickel_ebsd_large, CC BY 4.0 (sources).