How accurate is Spread Less?
Short answer - 91.4%.
The Long Answer
The correlation with the chance of getting the virus is based off of a study by the medical journal The Lancet. The Lancet is known as one of the world's oldest, best-known, and comprehensive peer-reviewed general medical journals.
"The primary limitation of our study is that all studies were non-randomised, not always fully adjusted, and might suffer from recall and measurement bias (eg, direct contact in some studies might not be measuring near distance). However, unadjusted, adjusted, frequentist, and Bayesian meta-analyses all supported the main findings, and large or very large effects were recorded. Nevertheless, we are cautious not to be overly certain in the precise quantitative estimates of effects, although the qualitative effect and direction is probably of high certainty."
Thus, while the data is not perfect by any means, it provides a good baseline to estimate the chance of getting the virus.
Our algorithm takes their risk findings and relates it with particle density, with particles about a certain threshold having a specified infection rate. Using the pixel values to evaluate the density, we were able to create an algorithm that matches closely with the research from The Lancet. The algorithm was created so that it recognizes that a person has around a 13% infection rate at 1 meter away, and a drastic reduction as the distance increases.
We found remarkable similarities between our graphs and The Lancet graphs, affirming the idea that density is an adequate method of finding the infection rate of the virus at various distances.
Here is the Lancet Study
The correlation with the chance of getting the virus and exposure time is based off a study by the University of the Philippines College of Medicine. Their study measured the time people were exposed to the virus and compared it to the number of newly infected people. The data used in our calculations was based on the exposure of talking to just one person, but the study also includes an 'encounter' value that could factor in the exposure of more than one person. The percent chance of being infected was based off of the fraction of one new infection in the graphs provided within the study.