Pedestron

Pedestron is a MMdetection based repository, that focuses on the advancement of research on pedestrian detection. We provide a list of detectors, both general purpose and pedestrian specific to train and test. Moreover, we provide pre-trained models and benchmarking of several detectors on different pedestrian detection datasets. Additionally, we provide processed annotations and scripts to process the annotation of different pedestrian detection benchmarks. If you use Pedestron, please cite us (see at the end) and other respective sources.

🔥 Updates 🔥

🧨 We have realeased PedesFormer - Transformer Based Pedestrian Detection repo (particularly Swin Transformer) along with pre-trained models. Stay tune for the updates. 🧨

YouTube demo

Caltech and EuroCity Persons. Pre-Trained model available.

Leaderboards

Installation

We refer to the installation and list of dependencies to installation file. Clone this repo and follow installation. Alternatively, Google Colab step-by-step instruction can be followed for installation (Please download the pre-trained models from the table in the readme.md, the link is broken on google colab for the pre-trained model). Addiitonally, you can also refer to the google doc file for step-by-step installation. For running a docker image please see installation file. Additionally, if you have more recent versions of Cuda (such as > 11.X), please have a look at the following PR request. There are some useful tips on how to install Pedestron, feel free to contribute.

List of detectors

Currently we provide configurations for the following detectors, with different backbones

Cascade Mask-R-CNN
Faster R-CNN
RetinaNet
RetinaNet with Guided Anchoring
Hybrid Task Cascade (HTC)
MGAN
CSP

Following datasets are currently supported

Datasets Preparation

We refer to Datasets preparation file for detailed instructions

Benchmarking

Benchmarking of pre-trained models on pedestrian detection datasets (autonomous driving)

Detector	Dataset	Backbone	Reasonable	Heavy
Cascade Mask R-CNN	CityPersons	HRNet	7.5	28.0
Cascade Mask R-CNN	CityPersons	MobileNet	10.2	37.3
Faster R-CNN	CityPersons	HRNet	10.2	36.2
RetinaNet	CityPersons	ResNeXt	14.6	39.5
RetinaNet with Guided Anchoring	CityPersons	ResNeXt	11.7	41.5
Hybrid Task Cascade (HTC)	CityPersons	ResNeXt	9.5	35.8
MGAN	CityPersons	VGG	11.2	52.5
CSP	CityPersons	ResNet-50	10.9	41.3
Cascade Mask R-CNN	Caltech	HRNet	1.7	25.7
Cascade Mask R-CNN	EuroCity Persons	HRNet	4.4	21.3
Faster R-CNN	EuroCity Persons	HRNet	6.1	27.0

Benchmarking of pre-trained models on general human/person detection datasets

Detector	Dataset	Backbone	AP
Cascade Mask R-CNN	CrowdHuman	HRNet	84.1

Getting Started

Running a demo using pre-trained model on few images

Pre-trained model can be evaluated on sample images in the following way

python tools/demo.py config checkpoint input_dir output_dir

Download one of our provided pre-trained model and place it in models_pretrained folder. Demo can be run using the following command

python tools/demo.py configs/elephant/cityperson/cascade_hrnet.py ./models_pretrained/epoch_5.pth.stu demo/ result_demo/

See Google Colab demo.

Training

single GPU training
multiple GPU training

Train with single GPU

python tools/train.py ${CONFIG_FILE}

Train with multiple GPUs

./tools/dist_train.sh ${CONFIG_FILE} ${GPU_NUM} [optional arguments]

For instance training on CityPersons using single GPU

python tools/train.py configs/elephant/cityperson/cascade_hrnet.py

Training on CityPersons using multiple(7 in this case) GPUs

./tools/dist_train.sh configs/elephant/cityperson/cascade_hrnet.py 7

Testing

single GPU testing
multiple GPU testing

Test can be run using the following command.

python ./tools/TEST_SCRIPT_TO_RUN.py PATH_TO_CONFIG_FILE ./models_pretrained/epoch_ start end\
 --out Output_filename --mean_teacher

For example for CityPersons inference can be done the following way

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

Alternatively, for EuroCity Persons

python ./tools/test_euroCity.py configs/elephant/eurocity/cascade_hrnet.py ./models_pretrained/epoch_ 147 148 --mean_teacher

or without mean_teacher flag for MGAN

python ./tools/test_city_person.py configs/elephant/cityperson/mgan_vgg.py ./models_pretrained/epoch_ 1 2\
 --out result_citypersons.json

Testing with multiple GPUs on CrowdHuman

./tools/dist_test.sh ${CONFIG_FILE} ${CHECKPOINT_FILE} ${GPU_NUM} [--out ${RESULT_FILE}] [--eval ${EVAL_METRICS}]

./tools/dist_test.sh configs/elephant/crowdhuman/cascade_hrnet.py ./models_pretrained/epoch_19.pth.stu 8 --out CrowdHuman12.pkl --eval bbox

Similarly change respective paths for EuroCity Persons
For Caltech refer to Datasets preparation file

Please cite the following work

CVPR2021

@InProceedings{Hasan_2021_CVPR,
    author    = {Hasan, Irtiza and Liao, Shengcai and Li, Jinpeng and Akram, Saad Ullah and Shao, Ling},
    title     = {Generalizable Pedestrian Detection: The Elephant in the Room},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2021},
    pages     = {11328-11337}
}

ArXiv2022

@article{hasan2022pedestrian,
  title={Pedestrian Detection: Domain Generalization, CNNs, Transformers and Beyond},
  author={Hasan, Irtiza and Liao, Shengcai and Li, Jinpeng and Akram, Saad Ullah and Shao, Ling},
  journal={arXiv preprint arXiv:2201.03176},
  year={2022}
}

thisishale/Pedestron