The entire project is based on https://github.com/longcw/faster_rcnn_pytorch and https://github.com/jwyang/faster-rcnn.pytorch
Tested on PASCAL VOC 2007, COCO 2017, Caltech Pedestrian Detection Benchmark.
Mainly pretrain multi-class detection model and initialize model that detects only pedestrians. I add triplet loss training impressed by Triplet-based Deep Similarity Learning for Person Re-Identification, ICCV Workshops 2017
paper : https://arxiv.org/abs/1802.03254 Person Re-ID problems organized in https://github.com/handong1587/handong1587.github.io/blob/master/_posts/deep_learning/2015-10-09-re-id.md
This code is built for individual study to detect Pedestrian dataset. Still need much work.
Thank you.