abs-hdf5 0.2.3

A Python + C++ toolkit for reading CAD B-Rep data from HDF5, sampling surfaces/curves, and generating blue-noise point clouds.

ABS-HDF5: Geometry processing & blue-noise sampling for HDF5 B-Rep data

STEP-to-HDF5 Conversion and Point Cloud Sampling (steptohdf5 + ABS-HDF5)

Introduction

This guide explains how to use steptohdf5 and ABS-HDF5 together to convert CAD models from STEP/STP files into HDF5 datasets, and then sample those datasets into point clouds. These two tools are designed to work in tandem:

• steptohdf5 – a converter that turns CAD solids (from *.step or .stp files) into analysis-ready HDF5 files. Each HDF5 file contains the model’s geometry (B-rep surfaces and curves), topology (faces, edges, etc.), and a high-quality triangular mesh for the surfaces. Currently, steptohdf5 is not available on PyPI or Conda, so it is run via a Docker container. (A PyPI/Conda release is planned – until then, use Docker as described below.)

• ABS-HDF5 – a Python/C++ toolkit (available on PyPI) that reads the HDF5 files produced by steptohdf5 and generates point cloud samples from the geometry. It provides command-line tools to export point clouds to PLY files or Python pickles, and also a Python API for advanced sampling (including fast Poisson-disk downsampling for blue-noise distributions).

Workflow Overview: Using these tools, a typical pipeline is:

1. Convert a CAD model from STEP to HDF5 (using steptohdf5).
2. Sample the HDF5 model to a point cloud (using ABS-HDF5).

This README will cover the installation of both tools, then walk through the conversion and sampling steps with examples (including single-file and batch processing), and provide basic API usage for each. Both projects are open-source under the Better Step initiative – see their GitHub repositories for more details:

• steptohdf5: https://github.com/better-step/cadmesh
• ABS-HDF5: https://github.com/better-step/abs

---

Installation

steptohdf5 (via Docker)

Since steptohdf5 is not yet on PyPI or Conda, the easiest way to use it is through its Docker image (which comes with all required dependencies, such as OpenCASCADE). Ensure you have Docker installed, then pull the steptohdf5 image from the registry:

docker pull itsmechandu/steptohdf5:latest

Note: steptohdf5 relies on OpenCASCADE (via pythonocc-core 7.4.0) for CAD B-rep processing and uses meshio for mesh generation. The Docker image has these pre-installed, so you don't need to install anything else on your host.

ABS-HDF5 (via pip)

ABS-HDF5 is distributed on PyPI. Install it into your Python environment (we recommend using a virtual environment):

pip install abs-hdf5

This installs the abs Python package along with two CLI tools: abs-to-ply and abs-to-pickle. No additional system dependencies are required.

Tip: Ensure your pip is up-to-date and you’re using Python 3.8 or newer.

---

Usage: Conversion and Sampling Pipeline

1. Convert STEP → HDF5 using steptohdf5

Run the Docker container with your input and output folders bind-mounted:

docker run --rm \
  -v /path/to/cad_workspace:/workspace \
  -w /workspace \
  itsmechandu/steptohdf5:latest \
  steptohdf5 <input.step> -o hdf5 -l logs

docker run --rm \
  -v /path/to/cad_workspace:/workspace \
  -w /workspace \
  itsmechandu/steptohdf5:latest \
  <input.step> \
  -o output \
  -l logs

• <input.step>: Path inside /workspace, e.g., cad_files/Model.step.
• -o hdf5: Output folder for .hdf5 files (inside /workspace).
• -l logs: Folder for log files.

Single-file example:

mkdir -p ~/cad_jobs/{cad_files,hdf5,logs}
cp MyModel.step ~/cad_jobs/cad_files/
cd ~/cad_jobs
docker run --rm \
  -v "$PWD":/workspace \
  -w /workspace \
  itsmechandu/steptohdf5:latest \
  steptohdf5 cad_files/MyModel.step -o hdf5 -l logs

After running, you’ll have:

~/cad_jobs/hdf5/MyModel.hdf5
~/cad_jobs/logs/MyModel.log

Batch conversion with a list:

ls cad_files/*.step > cad_files/list.txt
docker run --rm \
  -v "$PWD":/workspace \
  -w /workspace \
  itsmechandu/steptohdf5:latest \
  steptohdf5 --list cad_files/list.txt -o hdf5 -l logs -j 4

---

2. Sample HDF5 → Point Cloud using ABS-HDF5

With ABS-HDF5 installed, use the abs-to-ply CLI to generate PLY point clouds from HDF5:

abs-to-ply hdf5/MyModel.hdf5 samples -n 5000 -j 8

• hdf5/MyModel.hdf5: Input HDF5 file.
• samples: Output folder for PLY files.
• -n 5000: Points per part after Poisson-disk downsampling.
• -j 8: Parallel workers.

Batch PLY conversion:

Convert all HDF5 files in hdf5/:

abs-to-ply hdf5/ samples -n 3000 -j 8

This creates samples/MyModel_part001.ply, etc., for each part of each model.

abs-to-pickle example:

abs-to-pickle hdf5/ pickles -n 5000 -j 4

Generates .pkl files containing Python dicts:

{
  'file': 'MyModel.hdf5',
  'part': 1,
  'points': ndarray(N,3),
  'normals': ndarray(N,3)
}

---

API Usage

steptohdf5 Python API (cadmesh)

from steptohdf5.utils.processing import process_step_files

success, failed = process_step_files(
    input='cad_files/list.txt',
    output='/hdf5',
    log='/log')

ABS-HDF5 Python API (abs)

from  abs import read_parts, sample_parts

# Sample points + normals
def compute_labels(part, topo, points ):
  if topo.is_face(): return 1
  else : return 0

# Read parts from HDF5
parts = read_parts('hdf5/Model.hdf5')

P, S = sample_parts(parts, num_samples, compute_labels)

---

Development & Testing

# abs-hdf5
git clone https://github.com/better-step/abs.git
cd abs
pip install -e .[dev]
pytest -q

---

Contributing & License

• steptohdf5 (Python) – GPL-3.0
• abs-hdf5 Python bindings – MIT
• abs-hdf5 C++ core – MPL-2.0

Please review the Code of Conduct and open an issue before submitting larger changes.

---

Happy converting & sampling! – The Better Step maintainers