LAMICBench++ sample pipeline

April 19, 2026 ยท View on GitHub

If you want to reproduce the released LAMICBench++ numbers in bench/sample_scripts/README.md, start here.

cd bench/sample_scripts/lamicbench_plus
# unpack the released layout pack beside this script
tar -xzf lamicbench_layout_pack.tar.gz
# run with default arguments
python sample_bench.py

With defaults, generated images are written to bench/output/lamicbench_plus/images.

After generation, evaluate with:

python bench/lamicbench_plus/eval_lamicbench.py
python bench/lamicbench_plus/print_score.py

Main script and defaults

sample_bench.py defaults:

  • layout JSON: ./lamicbench_layout_pack/lamicbench_layout.json
  • layout images: ./lamicbench_layout_pack/layout_images/
  • reference images: bench/lamicbench_plus/lamicbench_plus_files/reference_images/
  • reference masks: bench/lamicbench_plus/lamicbench_plus_files/reference_masks/
  • output: bench/output/lamicbench_plus/

Model paths are read from CLI or env vars:

  • KONTEXT_MODEL_PATH
  • ADAPTER_PATH

Optional overrides

Use python sample_bench.py --help to override paths, seeds, width, batch size, and sample count.