Official code for "Drone-based RGBT Tiny Person Detection".
Please refer to https://github.com/open-mmlab/mmdetection/tree/2.x
mmdet 2.25.1
mmcv-full 1.6.1
pytorch 1.10.0
Models | mAP50 | mAP50(tiny) | mAP25 |
---|---|---|---|
QFDet | 42.08 | 44.04 | 57.34 |
QFDet* | 46.72 | 48.75 | 61.62 |
Models | mAP | mAP50 | mAP75 |
---|---|---|---|
QFDet | 31.10 | 70.40 | 22.90 |
QFDet* | 33.30 | 75.50 | 24.20 |
Train QFDet on RGBTDronePerson.
python tools/train.py qfdet_configs/qfdet_r50_fpn_1x_rgbtdroneperson.py
Train QFDet* on RGBTDronePerson.
python tools/train.py qfdet_configs/qfdet_star_r50_fpn_1x_rgbtdroneperson.py
Train QFDet on VTUAV-det.
python tools/train.py qfdet_configs/qfdet_r50_fpn_1x_rgbtdroneperson.py
Train QFDet* on VTUAV-det.
python tools/train.py qfdet_configs/qfdet_star_r50_fpn_1x_rgbtdroneperson.py
For example, test checkpoint epoch_11_qfdet_rgbtdroneperson.pth:
python tools/test.py qfdet_configs/qfdet_r50_fpn_1x_rgbtdroneperson.py work_dir/qfdet_r50_fpn/rgbtdroneperson/epoch_11_qfdet_rgbtdroneperson.pth --eval bbox
Please refer to our github page.
@article{ZHANG202361,
title = {Drone-based RGBT tiny person detection},
journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
volume = {204},
pages = {61-76},
year = {2023},
doi = {https://doi.org/10.1016/j.isprsjprs.2023.08.016},
url = {https://www.sciencedirect.com/science/article/pii/S0924271623002319},
author = {Yan Zhang and Chang Xu and Wen Yang and Guangjun He and Huai Yu and Lei Yu and Gui-Song Xia}
}