Skip to content
LayerDynamics — LatticeLabs Toolkit LayerDynamics — LatticeLabs Toolkit LatticeLabs Toolkit

ll_brepnet — Usage

The pipeline is STEP → .npz records → train → segment. Everything below uses the real Fusion 360 Gallery segmentation dataset (s2.0.0).

1. Get the dataset

Section titled “1. Get the dataset”
Terminal window
curl https://fusion-360-gallery-dataset.s3-us-west-2.amazonaws.com/segmentation/s2.0.0/s2.0.0.zip -o s2.0.0.zip
unzip s2.0.0.zip   # → s2.0.0/{breps/step, breps/seg, train_test.json, segment_names.json}

2. Prepare records + manifest

Section titled “2. Prepare records + manifest”

quickstart extracts each STEP file to a .npz, copies the matching .seg labels, and writes a dataset.json (official train/test split, train-only feature standardization, the 8 class names). Use --train-subset / --test-subset to work on a subset, or omit them for the full 35,680 solids.

Terminal window
python -m ll_brepnet.pipelines.quickstart \
  --dataset-dir ./s2.0.0 \
  --output-dir  ./processed \
  --num-workers 12 \
  --train-subset 4000 --test-subset 800

3. Train

Section titled “3. Train”
Terminal window
python -m ll_brepnet.train \
  --dataset-file ./processed/dataset.json \
  --dataset-dir  ./processed \
  --label-dir    ./processed \
  --max-epochs 30 --num-layers 4 --hidden-dim 128 \
  --learning-rate 0.002 --accelerator auto --log-dir ./logs

Checkpoints (best by validation mIoU) and TensorBoard logs land under --log-dir. --accelerator auto uses an Apple-Silicon GPU (MPS) or CUDA when available. On the full official split (no --*-subset), this produced test mIoU ≈ 0.828 / accuracy ≈ 0.947 on the 5,366-solid test split (see Overview).

4. Segment STEP files

Section titled “4. Segment STEP files”
Terminal window
python -m ll_brepnet.eval.evaluate \
  --step-dir ./s2.0.0/breps/step \
  --model ./logs/ll_brepnet/version_0/checkpoints/best.ckpt \
  --dataset-file ./processed/dataset.json \
  --output-dir ./predictions

Each input solid gets a <stem>.logits file: one row per face, one column per class. The predicted segment of face i is the argmax of row i.

Python API

Section titled “Python API”
from ll_brepnet import (
    extract_brepnet_data_from_step,  # STEP folder → .npz records
    prepare_fusion360,               # Fusion 360 dir → records + dataset.json
    BRepDataModule,                  # LightningDataModule
    LLBRepNet,                       # the model (a LightningModule)
    evaluate_folder,                 # checkpoint + STEP folder → per-face logits
)


# Inference from an existing checkpoint:
written = evaluate_folder(
    step_dir="./s2.0.0/breps/step",
    model_ckpt="./logs/.../best.ckpt",
    dataset_file="./processed/dataset.json",
    output_dir="./predictions",
    num_workers=8,
)

The model also exposes predict_logits(batch) → [(file_stem, per-face softmax)] for programmatic, per-solid inference.

Alternate input: precomputed JSON topology

Section titled “Alternate input: precomputed JSON topology”

If you already have the coedge topology + features from another tool, skip the OpenCascade extraction and feed JSON directly:

Terminal window
python -m ll_brepnet.pipelines.extract_brepnet_data_from_json \
  --json my_solid.json --output my_solid.npz

See BRepJsonExtractor for the schema (coedge incidence arrays + per-face/edge feature matrices; UV-grids optional).