Keyence-StitchEDOF

stitchedof.py performs two steps for tiled Keyence datasets:

1. Per-tile extended depth-of-field (EDOF) collapse from Z-stacks.
2. Global tile stitching into one output TIFF.

The script expects an input folder containing:

• Tile TIFF files named like ..._00001_Z001.tif or ..._00001_Z001_CH4.tif
• Exactly one .gci file (used to read row/column grid size)

Requirements

• Python packages: numpy, tifffile, scipy, tqdm, torch
• CUDA-enabled PyTorch is optional but recommended for faster phase-correlation stitching.

PyTorch install:

• https://pytorch.org/get-started/locally/

Usage

python stitchedof.py <input_dir> [output]

• input_dir: required directory containing TIFF tiles and one .gci
• output: optional output TIFF path or filename
  • If omitted, default output is: <input_dir>/<input_dir_name>_stitched.tif

Examples:

python stitchedof.py data
python stitchedof.py data custom_output.tif
python stitchedof.py data /tmp/out/final.tif --workers 4 --overlap 0.3
python stitchedof.py data --tilestitch

Options

• --patch_size N EDOF local-statistics window size (default: 7)
• --alpha A EDOF score mixing factor in [0,1] (default: 0.7)
• --workers N multiprocessing workers for EDOF tiles (default: CPU count)
• --edofsave BOOL save per-tile EDOF TIFFs (true/false, default: false)
• --overlap F fractional tile overlap for alignment/blending (default: 0.3)
• --tilestitch use legacy single-tile row anchoring instead of default full-row overlap stitching

Notes

• Default stitching mode is row-based (ROWSTITCH), which aligns rows using full overlap strips.
• --tilestitch switches to legacy behavior using one anchor tile pair per row.
• Runtime console output reports stitching mode and device (CPU or CUDA).