Awesome
F2DNet
<img title="Frankfurt" src="gifs/gm.png" width="800" />F2DNet is a Pedestron based repository which implements a novel, two-staged detector i.e. Fast Focal Detection Network for pedestrian detection.
<img title="Frankfurt" src="gifs/1.gif" width="400" /> <img title="Frankfurt" src="gifs/2.gif" width="400"/>
Installation
Please refer to base repository for step-by-step installation.
List of detectors
In addition to configuration for different detectors provided in base repository we provide configuration for F2DNet.
Following datasets are currently supported
Datasets Preparation
Please refer to base repository for dataset preparation.
Benchmarking
Benchmarking of F2DNet on pedestrian detection datasets
| Dataset | ↓Reasonable | ↓Small | ↓Heavy |
|---|---|---|---|
| CityPersons | 8.7 | 11.3 | 32.6 |
| EuroCityPersons | 6.1 | 10.7 | 28.2 |
| Caltech Pedestrian | 2.2 | 2.5 | 38.7 |
Benchmarking of F2DNet when trained using extra data on pedestrian detection datasets
| Dataset | Config | Model | ↓Reasonable | ↓Small | ↓Heavy |
|---|---|---|---|---|---|
| CityPersons | cascade_hrnet | Cascade Mask R-CNN | 7.5 | 8.0 | 28.0 |
| CityPersons | ecp_cp | F2DNet | 7.8 | 9.4 | 26.2 |
| Caltech Pedestrian | cascade_hrnet | Cascade Mask R-CNN | 1.7 | 25.7 | |
| Caltech Pedestrian | ecp_cp_caltech | F2DNet | 1.7 | 2.1 | 20.4 |
References
Please cite the following work
@inproceedings{khan2022f2dnet,
title={F2DNet: fast focal detection network for pedestrian detection},
author={Khan, Abdul Hannan and Munir, Mohsin and van Elst, Ludger and Dengel, Andreas},
booktitle={2022 26th International Conference on Pattern Recognition (ICPR)},
pages={4658--4664},
year={2022},
organization={IEEE}
}