Utilizing Weak Supervision to Infer Complex Objects and Situations in Autonomous Driving Data

While the detection and classification of simple objects encountered during autonomous driving sessions has been widely researched, the detection of complex objects and situations based on the combinations of objects in a scene remains relatively overlooked. This is especially difficult due to the cost of gathering labels for each complex scenario of interest before training a specialized model. To address this bottleneck of training data, we explore the applicability of weak supervision, or relying on higher level, noisier forms of supervision to label training data. Specifically, we use data programming, a paradigm that can learn the accuracy and dependency structure of these sources without using any ground truth labels and assign training labels accordingly. We focus on an example task of cyclist detection by comparing weak supervision, which relies on a set of user-defined rules over the outputs of detectors that identify people and bikes separately, to CyDet [1], which detects the cyclist as a complete object. We find that the weak supervision method can achieve a performance of 96.8 F1 points, 4.6 F1 higher than CyDet, without relying on any ground truth labels on the newly released Specialized Cyclist Dataset. We then discuss how heuristics can detect complex objects such as cyclists and by extension, situations, based on the output of existing object detection algorithms.