dmy1997 / DHRNet

Dynamic HRNet for pedestrian detection

DHRNet

DHRNet aims to generate different scale-sensitive weights (a parallel multi-branch architecture with a soft conditional gate module) during feature fusion in HRNet, thus consistently improving the performance across different scales on Citypersons.

Core files

The code base of our work is MMDetection(Here is MMDetection-1.2.0).
Config: configs/hrnet/faster_rcnn_hrnetv2p_w18_dynamic.py
Backbone: DHRNet

Citation

@article{DHRNet,
  title   = {Learning a Dynamic High-Resolution Network for Multi-Scale Pedestrian Detection.},
  author  = {Mengyuan Ding, Shanshan Zhang, Jian Yang.},
  journal = {International Conference on Pattern Recognition(ICPR)},
  year={2020}
}