Before you start know this - this is not perfect. I've been able to get quality matches around 25-30% of the time. Here is how it works. Images are first checked for EXIF meta data - if it's there job done. If not, 2 different convolutional neural networks are used that we're trained on location photos from Wikipeida and other data sets easy to categorize data sets. If people use it - I'll spend more time making it awesome
2-13 Added Landmark and object detection