Varying Adaptive Ensemble of Deep Detectors for Road Damage Detection

Road damage detection: A task of identifying damages in roads and categorizing the same has been explored in recent times due to availability of multiple public data sets. In this paper, we present our system that was developed as part of the IEEE Big Data Cup 2018. We developed a varying adaptive ensemble of multiple R-CNN based deep learning detectors for each of the seven cities. We developed and experimented with multiple configuration of the detector models with varying pretraining criterion and achieved a best accuracy score of 0.60 respectively for an adaptive ensemble of seven R-CNN models pretrained over KITTI object detection benchmark.