A Study on Manhole Detection and Classification with Deep-learning based Methods

抄録

On this report, we would be presenting our observation on various object detection methods on manhole dataset. The first method that we tried is the original Single Shot Detector built on top of Caffe; we tried both input size available 300x300 and 512x512. For both experiments, we got pretty promising results. In addition to those experiments, we also tried Single Shot Detector built on Keras and did basically the same experiment. The result is not that great in comparison to SSD on Caffe.