We present a method to automatically segment indoor scenes by detecting repeated objects. Our algorithm scales to datasets with 198 million points and does not require any training data. We propose a trivially parallelizable ...