<ul><li>Researchers develop a method to digitize historical housing assessment data to quantify financial effects of 1930s U.S. government housing policies on racial wealth disparities.</li><li>They manually annotated property cards for over 12,000 properties and used OCR to label data for an additional 50,000 properties.</li><li>A two-stage approach combining classical computer vision techniques with deep learning-based OCR was utilized in the process.</li><li>A regression model based on building feature data was also employed to estimate historical values in cases where OCR could not be applied, showing promise for generalizability to other counties.</li></ul>

Predicting the Past: Estimating Historical Appraisals with OCR and Machine Learning

Discover more