As part of the course CS4245 Seminars Computer Vision by Deep Learning at the Delft University of Technology, we have developed a pipeline in order to convert sketches of electronic circuits into a labeled image. For a detailed overview of this project, please visit the blogpost.
First create a virtual environment and install dependencies (tested for Python 3.10.12)
git clone git@github.com:Timdnb/electronic-component-classification.git
cd electronic-component-classification
python -m venv venv
source venv/bin/activate
pip install -r requirements.txt
To perform inference on your electronic circuit sketch(es), run the following command:
python detect.py --img-path <single-img | img-directory> [--outdir <outdir>]
Arguments:
--img-path: can be both a single image or a directory containing images--outdir: path where you want the output to be stored, defaults to the same directory as the input image(s)
After running a window will appear, in this window use the sliders to best extract the lines of your electronic circuit. Once you are satisfied press "Enter" to continue.
In the current state the pipeline classifies all components and junctions, however it does not yet convert it to a digital counterpart. This can be a future improvement.
Original:
Labeled:
The notebooks/ folder contains all notebooks that were used during development, they can be investigated to have a better look into our methods. However one of the notebooks requires the presence of a hand-drawn electronic component dataset from kaggle. Download this dataset and put the folders of the separate components under a dataset/components/ directory (make this directory yourself).