A new system has been developed that can efficiently search through hundreds of hours of video for events and visual features that correspond to a few training examples. The researchers trained their system to recognize different stages of an operation, such as biopsy, tissue removal, stapling, and wound cleansing.