Data Collection and Annotation

Building CV, AR and VR systems for unconstrained applications require a large amount of labeled data. The availability of correctly labeled and validated data is a major bottleneck in building accurate trustable systems. For several problems, classification engines can be designed by performing convenience sampling and collecting data from the web. However, building trustable and explainable systems require annotations not only with respect to the output class, but also with respect to attributes, semantic segmentation, and marking the regions of importance. While data annotation for object recognition can be outsourced, annotation for specialized applications like the ones in the healthcare vertical requires close collaboration with domain experts. To facilitate this activity, we plan to set up a data collection and annotation lab.