SNIP_Pedestrian-Detection

Introduction

This project is to re-implement the top-performing multi-scale method for generic object detection SNIP on the pedestrian detection task. We use Adapted Faster RCNN as the base detector on CityPerson dataset, which achieves remarkable improvement using a multi-scale test setting. We add ignore region handling into a faster pytorch implementation of faster R-CNN to implement the Adapted Faster RCNN. The code will be released soon.