Bin#

Calibration-aware 4D-STEM binning widget with BF/ADF quality checks.

Usage#

import numpy as np
from quantem.widget import Bin

data = np.random.rand(64, 64, 128, 128).astype(np.float32)
w = Bin(data, pixel_size=2.39, k_pixel_size=0.46, device="cpu")
w

Features#

Torch-only compute with configurable device (cpu/cuda/mps)
Independent scan + detector bin factors with mean/sum reduction
Edge handling modes: crop, pad, error
Calibration propagation for real-space and detector sampling after binning
BF/ADF QC previews before and after binning with compact stats
Export helpers: save_image, save_zip, save_gif

Methods#

w = Bin(data)

# Replace data (preserves bin settings)
w.set_data(new_data)

# Access results
w.result              # binned tensor (on device)
w.get_binned_data()   # copy as tensor
w.get_binned_data(as_numpy=True)  # copy as NumPy

State Persistence#

w = Bin(data, pixel_size=2.39, k_pixel_size=0.46)

w.summary()                # human-readable summary
w.state_dict()             # serializable state dictionary
w.save("bin_preset.json")  # save preset JSON

w2 = Bin(data, state="bin_preset.json")

Programmatic Export#

w.save_image("bin_grid.png", view="grid")
w.save_zip("bin_bundle.zip", include_arrays=True)
w.save_gif("bin_bf_compare.gif", channel="bf")

Batch Runner#

Use the same preset over folders (one file at a time):

python -m quantem.widget.bin_batch \
  --input-dir /path/to/raw \
  --output-dir /path/to/binned \
  --preset bin_preset.json \
  --pattern '*.npy' \
  --recursive \
  --device cpu

Examples#

API Reference#

See quantem.widget.Bin for full documentation.