About PlankAssembly

PlankAssembly is a research software presented at ICCV 2023 for robust 3D reconstruction from three orthographic views using learnt shape programs. Developed by researchers including Wentao Hu and Jia Zheng, the tool transforms engineering drawings into 3D geometric models by leveraging a program-based representation. The system supports multiple input configurations, including visible and hidden lines, visible edges only, and sidefaces, enabling flexible processing of technical diagrams. It is implemented in Python using PyTorch and PyTorch Lightning, with dependencies on PythonOCC for rendering and geometric operations. The software includes a dataset for training and testing, which can be rendered into three-view orthogonal drawings or processed as SVG and JSON files. Users can also explore variations such as raster image inputs or comparisons with PolyGen through dedicated repository branches. PlankAssembly is particularly useful for applications in computer vision, CAD automation, and 3D modeling where p

m

Published by

manycore-research

Visit View Profile

README.md

View on GitHub

PlankAssembly: Robust 3D Reconstruction from Three Orthographic Views with Learnt Shape Programs

Wentao Hu* · Jia Zheng* · Zixin Zhang* · Xiaojun Yuan · Jian Yin · Zihan Zhou

IEEE/CVF Conference on Computer Vision (ICCV), 2023

*These authors contributed equally to this work.

[!NOTE] In our follow-up work, CAD2Program, we discovered that a modern vision models (e.g., ViT) can understand engineering drawings. For detailed implementation, please check vit branch.

[!NOTE] This branch contains the implementation of PlankAssembly, which supports three types of inputs: (1) visible and hidden lines, (2) visible edges only, and (3) sidefaces. For raster images as inputs, please refer to the raster branch. For comparison with PolyGen, please refer to the polygen branch.

Setup

Our code has been tested with Python 3.8, PyTorch 1.10.0, CUDA 11.3, and PyTorch Lightning 1.7.6.

Installation

Clone the repository, then create and activate a plankassembly conda environment using the following commands.

# clone repository
git clone https://github.com/manycore-research/PlankAssembly.git
# create conda env
conda env create --file environment.yml
conda activate plankassembly

If you encounter any issue with provided conda environment, you may install dependencies manually using the following commands.

conda create -n plankassembly python=3.8
conda activate plankassembly
conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=11.3 -c pytorch -c conda-forge
pip install pytorch-lightning==1.7.7 torchmetrics==0.11.4 rich==12.5.1 'jsonargparse[signatures]'
pip install detectron2 -f https://dl.fbaipublicfiles.com/detectron2/wheels/cu113/torch1.10/detectron2-0.6%2Bcu113-cp38-cp38-linux_x86_64.whl
conda install -c conda-forge pythonocc-core=7.6.2
pip install numpy shapely svgwrite svgpathtools trimesh setuptools==59.5.0 html4vision

Dataset

The dataset can be found on Hugging Face Datasets. Please download the data first, then unzip the data in the project workspace.

The released dataset only contains 3D shape programs. To prepare the data for training and testing, please run the following commands.

We use PythonOCC to render three-view orthogonal engineering drawings and save them as SVG files.

# render complete inputs
python dataset/render_complete_svg.py
# render noisy inputs, please specify the noise ratio
python dataset/render_noisy_svg.py --data_type noise_05 --noise_ratio 0.05
# render visible inputs
python dataset/render_visible_svg.py

Then, pack the input line drawings and output shape programs into JSON files.

python dataset/prepare_info.py --data_path path/to/data/root

To visualize the 3D model, we could generate the ground-truth 3D meshes from shape.

python misc/build_gt_mesh.py --data_path path/to/data/root

Training

Use the following command to train a model from scratch:

# train a model with complete lines as inputs
python trainer_complete.py fit --config configs/train_complete.yaml

Testing

Use the following command to test with a pre-trained model:

# infer a model with complete lines as inputs
python trainer_complete.py test \
    --config configs/train_complete.yaml \
    --ckpt_path path/to/checkpoint.ckpt \
    --trainer.devices 1

Evaluation

To compute the evaluation metrics, please run the following command:

python evaluate.py --data_path path/to/data/dir --exp_path path/to/lightning_log/dir

Visualization

To visualize the results, we build 3D mesh models from predictions:

python misc/build_pred_mesh.py --exp_path path/to/lightning_log/dir

Then, we use HTML4Vision to generate HTML files for mesh visualization (please refer to here for more details):

python misc/build_html.py --exp_path path/to/lightning_log/dir

The 2D images presented in our paper are rendered using bpy-visualization-utils.

Model Checkpoints

The checkpoints can be found on Hugging Face Models. Or click the links below to download the checkpoint for the corresponding model type directly.

Model trained on complete inputs: here
Model trained on visible inputs only: here
Model trained on sideface inputs: here

LICENSE

PlankAssembly is licensed under the AGPL-3.0 license. The code snippets in the third_party folder are available under Apache-2.0 License.

PlankAssembly