Pedestrian Detection: The Elephant In The Room Irtiza Hasan 1 , Shengcai Liao 1 , Jinpeng Li 1 , Saad Ullah Akram 2 , and Ling Shao 1 1 Inception Institute of Artificial Intelligence (IIAI), Abu Dhabi, UAE {irtiza.hasan,shengcai.liao,jinpeng.li,ling.shao}@inceptioniai.org 2 Department of Computer Science, Aalto University, Finland saad.akram@aalto.fi Abstract. Pedestrian detection is used in many vision based applica- tions ranging from video surveillance to autonomous driving. Despite achieving high performance, it is still largely unknown how well ex- isting detectors generalize to unseen data. To this end, we conduct a comprehensive study in this paper, using a general principle of direct cross-dataset evaluation. Through this study, we find that existing state- of-the-art pedestrian detectors generalize poorly from one dataset to an- other. We demonstrate that there are two reasons for this trend. Firstly, they over-fit on popular datasets in a traditional single-dataset training and test pipeline. Secondly, the training source is generally not dense in pedestrians and diverse in scenarios. Accordingly, through experiments we find that a general purpose object detector works better in direct cross-dataset evaluation compared with state-of-the-art pedestrian detec- tors and we illustrate that diverse and dense datasets, collected by crawl- ing the web, serve to be an efficient source of pre-training for pedestrian detection. Furthermore, we find that a progressive training pipeline works good for autonomous driving oriented detector. We improve upon previ- ous state-of-the-art on reasonable /heavy subsets of CityPersons dataset by 1.3%/1.7% and on Caltech by 1.8%/14.9% in terms of log average miss rate (MR -2 ) points without any fine-tuning on the test set. Detector trained through proposed pipeline achieves top rank at the leaderborads of CityPersons [42] and ECP [4]. Code and models will be available at https://github.com/hasanirtiza/Pedestron. Keywords: Pedestrian Detection, Autonomous Driving, Video Surveil- lance, Transfer Learning and Robotics 1 Introduction Pedestrian detection is a very actively researched task in the computer vision community, both in academia and industry. It has applications in many different domains, including robotics and autonomous vehicles, entertainment, and smart Corresponding author arXiv:2003.08799v2 [cs.CV] 22 Mar 2020