Discovering Design Concepts for CAD Sketches

This repository contains the core implementation of our paper:

[Discovering Design Concepts for CAD Sketches](NeurIPS 2022) <br> Yuezhi Yang, Hao Pan <br> <br>

Installation

Following is the suggested way to install the dependencies of our code:

git clone --recursive https://github.com/yyuezhi/SketchConcept.git

cd SketchConcept

conda create  -n sketch_concept python=3.8

conda activate sketch_concept

conda install pytorch==1.13.1 lz4 pytorch-cuda=11.6 matplotlib=3.7.1 -c pytorch -c nvidia

pip install -e SketchGraphs

pip install tensorboard scipy networkx numpy==1.23.4

Data Preparation

We provide preprocessed SketchGraph dataset and corresponding index file which filters the sketches we used to train the model. Please download the pretrained models from Google Drive and put them in Data/. For file size issue, the data are splited into two patches of roughly equal size.

Pre-trained Models

We provide the pretrained models for design intent interpretation. Please download the pretrained models from Google Drive and put them in experiment_directory/sample_pretrain_exp/result/network_param.

Usage

Train from Scratch

python train.py -d ./Data/ -e experiment_directory/sample_exp  --train_file sketch1_randnorm_sorted.npy

Generate sketch result for visualization using pretrain checkpoints

python train.py -d ./Data/ -e experiment_directory/sample_pretrain_exp  --train_file sketch1_randnorm_sorted.npy -v True -c 180

Visualize sketch result interactively

python plot_interactive_visualize.py -d ./Data/ -e <experiment_directory>  --train_file <train_file>  --epoch <epoch_num>

Visualize sketch result in library perspective

python plot_library_visualize.py -d ./Data/ -e <experiment_directory>  --train_file <train_file>   --epoch <epoch_num>

Visualize SketchGraph

python make_graph.py -d ./Data/ -e <experiment_directory>  --train_file <train_file>   --epoch <epoch_num>

Notice that this program need installation of PyGraphViz. Please install it according to your own environment.

Citation

If you find our work useful in your research, please consider citing:

@inproceedings{yang2022sketchconcept,
 author = {Yuezhi Yang, Hao Pan},
 booktitle = {Advances in Neural Information Processing Systems},
 title = {Discovering Design Concepts for CAD Sketches},
 volume = {35},
 year = {2022}
}