We are currently updating this repository. We are preparing training and testing launchers, which will be pushed soon. We have added a pre-trained model for CityPersons.
Pedestron is an MMetection based repository that focuses on the advancement of reserach on pedestrian detection. We provide processed annotations and scripts to process the annotation of different pedestrian detection benchmarks.
We refer to the installation and list of dependencies installation file. Clone this repo and follow installation of mmdetection.
*Currently we provide configurations for Cascade Mask-RCNN
-
Download the datasets from the official sites. Fill in the copyright forms where applicable.
-
Place them in ./datasets folder in the follwoing heararchy, for example annotation file for CityPersons should be (./datsets/CityPersons/) and images should be (./datsets/CityPersons/leftImg8bit_trainvaltest/leftImg8bit/train) for training images and collapse all validtaion images into(./datsets/CityPersons/leftImg8bit_trainvaltest/leftImg8bit/val_all_in_folder/).
-
Use our pre-processing script to convert the annotations into Pedestron acceptable format.
- CityPersons
- Caltech [link]
- EuroCity Persons [link]
- Download the pretrained CityPersons model and place it in the folder "models_pretrained/".
- Run the following command:
python ./tools/test_city_person.py configs/elephant/cityperson/cascade_hrnet.py ./models_pretrained/epoch_ 5 6\
--out result_citypersons.json --mean_teacher Coming Soon
Coming Soon
@article{irtiza20elephant,
title={Pedestrian Detection: The Elephant In The Room},
author={Hasan, Irtiza and Liao, Shengcai and Li, Jinpeng and Akram, Saad Ullah and Ling, Shao},
journal={arXiv preprint},
year={2020}}