I’m entering the final stages of a figure search engine, a nice wrapper for the new API method discussed below. It’s also a chance to properly release data mined directly from arxiv figures, and take advantage of the lambda + S3 processing pipeline I developed when pushing the p2t algorithms to cloud initially. Attached is an image showing a successful dynamoDB table entry, after upload of an input pdf to an endpoint, and processing by the pipeline. The schema you see will allow for search on figure labels (ie. measurement units), axis data ranges, models for the data represented in the figures themselves, as well as input and output images extracted directly from the document pages using machine learning.