This repository provides the implementation code for “Don’t Crosstalk to Me: Origami Structure-Augmented Sensing for Scalable Surface Pressure Monitoring” Sensys 24
To get the end-to-end results clone this repository and run the "setup_docker.sh" script. The script uses a docker environment to run the inference code using the pre-trained models provided in "../artifact" folder to run the inference on real-world data provided in "../figure_7x11/realworld_data2" and save the output results in "../output" as "overall_results.png"
git clone https://github.com/sr004/omsense.git
cd omsense
./setup_docker.sh"../3D model" folder contains the STL files required for printing the physical surface augmentation structure.
The main source code for training and inference can be found in the "../artifact folder".
Run "CG_CNN_training.py" file to train a new model from scratch, this script uses the simulation data provided "../figure_7x11".
Inference code is provided in "CG_CNN_inference.py", the docker command above uses this script to generate the overall results.
We also provide the pre-trained models for OMSense "model_5.pth" and velostat baseline "velostat_model_5.pth".
We also provide the simulation and real-world data used for this work for OMSense in "../figure_7x11", we simulate data for multiple configurations based on the resistance of conductors used in manufacturing.
The simulation dataset for different configurations is organized based on conductor resistance between two consecutive sensors in the matrix. "r_top" represents resistance value of the top conductor, "r_bottom" represents resistance of bottom conductor in ohm.
We use simulation data in "../figure_7x11/r_top=3.0,r_bottom=0.02/" data for training our model.
The real-world data used for testing for both OMSense and Velostat baseline can be found in "../figure_7x11/realworld_data2"
2 GB RAM and 4GB of hard disk space.
• Ubuntu-20.04 • Docker-27.2.0 • Conda
contact: srohal@ucmerced.edu