Google is mapping visually the world offering services like Street View on Google Maps and now Lens on Google Photos. This month, Google released a dataset with more than 30 thousand landmarks, with more than 2 million images. According to André Araujo and Tobias Weyand, Software Engineers at Google Research, the release has “a number of classes that are around 30 times larger than what is available in commonly used datasets.” This is the largest worldwide dataset for recognition of human-made and natural landmarks.
The release will allow researchers to have a dataset to work on, “in order to continue advancing the state of the art in computer vision.” Until now, “a significant obstacle for research in this area has been the lack of large annotated datasets.”
The challenge for Google is the recognition of the correct landmarks in a dataset of challenging test images. “Even within a large annotated dataset, there might not be much training data available for some of the less popular landmarks.”