You are provided with a total of ~1,000 cadastral survey plans: ~700 training samples and ~300 test samples.

Each training sample includes a scanned cadastral plan (PDF or image format), associated metadata such as lot number and surveyor name and the corresponding polygon defining the land parcel provided in .shp format.

Your task is to develop a model or workflow that can extract the same information—both the polygon and the metadata fields—from the 300 test samples, which only include the scanned plans.