j-IR-vis: Vision model for Infrared Spectroscopy (IR) embeddings.

This repository contains the code for reproducing the results for j-IR-vis, or to run inference on custom inputs.

Modelling

The below diagram illustrates the j-IR-vis pipeline. The input is the raw IR spectrocopy plot, and the output is a Multi Label Classification of the functional groups. The underlying model inspired by the Resnet family of vision models.

Dataset

Sample Dataset

A sample dataset is included with this repository for preliminary use of jirvis. This dataset contains 32 random IR Spectroscopy images along with their functional group labels from both the simulated and experimental datasets.

Custom Dataset

See this tutorial for more information on how to incorporate your own custom dataset.

Installation

1. Clone the Repository

git clone https://github.com/ChemAI-Lab/jirvis.git
cd jirvis

2. Create and Activate the Conda Environment

conda env create -f environment.yaml
conda activate jirvis

Using j-IR-vis

Adjust config files or use command line overrides as below.

j-IR-vis Training

Using the experimental IR dataset.

python scripts/train.py data=exp_ir

To use the simulated dataset, use data=exp_ir instead.

Running Inference

python scripts/inference.py

Reference

@article{jirvis,
title={j-IR-vis: Vision model for Infrared spectroscopy embeddings},
DOI={10.26434/chemrxiv-2025-d0j2v},
journal={ChemRxiv},
author={Sondhi, Rudra and Chacko, Edwin and Vargas-Hernández, Rodrigo A.},
year={2025}
}