In this paper, we present Breadcrumbs, a mobility dataset collected in the city of Lausanne (Switzerland) from multiple mobile phone sensors (GPS, WiFi, Bluetooth) from 81 users for a duration of 12 weeks. Currently available mobility datasets are restricted to geospatial information obtained through a single sensor at low spatiotemporal granularities. Furthermore, this passively collected data lacks ground-truth information regarding points of interest and their semantic labels. These features are critical in order to push the possibilities of geospatial data analysis towards analyzing mobility behaviors and movement patterns at a fine-grained scale. To this end, Breadcrumbs provides ground-truth and semantic labels for the points of interest of all the participants. The dataset also contains fine-grained demographic attributes, contact records, calendar events and social relationship tags between the participants. In order to demonstrate the significance of the ground-truth annotations, we discuss several use cases of this dataset. Furthermore, we compare four contrasting and widely used unsupervised clustering approaches for point of interest extraction from geolocation trajectories. Using the ground-truth information, we perform a detailed performance validation of these techniques and highlight their strengths and weaknesses. Given that mobility data is derived from an individuals inherent need of participating in activities, narrowing the gap between raw trajectory data points and complete trip annotation in essential. We thus make Breadcrumbs accessible to the research community in order to facilitate research in the direction of supervised human mobility learning schemes.