Awesome

UFPMP-Det: Toward Accurate and Efficient Object Detection on Drone Imagery

The repo is the official implementation of UFPMP-Det.

The code of UFP module is at mmdet/core/ufp

The code of MP-Det is at mmdet/models/dense_heads/mp_head.py

The config of our project is at configs/UFPMP-Det

Install

1. This repo is implemented based on mmdetection. Please install it according to get_start.md.

2. pip install nltk
   pip install albumentations
   

Quickstart

We provide the Dataset(COCO Format) as follows:

• VisDrone:链接：https://pan.baidu.com/s/1FfAsAApHZruucO5A2QgQAg 提取码：qrvs
• UAVDT:链接：链接：https://pan.baidu.com/s/1KLmU5BBWwgtFbuZa7QWavw 提取码：z08x

We provide the checkpoint as follows:

• VisDrone Coarse-Det:链接: https://pan.baidu.com/s/1jK3bqImDGSwqRJGVXinS0w 提取码: nab3
• VisDrone MP-Det ResNet50: 链接: https://pan.baidu.com/s/1zOoJVO2fPejnzM9KioZLuQ 提取码: m7rj

Training

This repo is only supposed single GPU.

Prepare

Build by yourself: We provide two data set conversion tools.

# conver VisDrone to COCO
python UFPMP-Det-Tools/build_dataset/VisDrone2COCO.py
# conver UAVDT to COCO
python UFPMP-Det-Tools/build_dataset/UAVDT2COCO.py
# build UFP dataset(VisDrone)
CUDA_VISIBLE_DEVICES=2 python UFPMP-Det-Tools/build_dataset/UFP_VisDrone2COCO.py \
    ./configs/UFPMP-Det/coarse_det.py \
    ./work_dirs/coarse_det/epoch_12.pth \
    xxxxxx/dataset/COCO/images/UAVtrain \
    xxxxxx/dataset/COCO/annotations/instances_UAVtrain_v1.json \
    xxxxxx/dataset/COCO/images/instance_UFP_UAVtrain/ \
    xxxxxx/dataset/COCO/annotations/instance_UFP_UAVtrain.json \
    --txt_path path_to_VisDrone_annotation_dir

Download:

In Quick Start

Train Coarse Detector

CUDA_VISIBLE_DEVICES=0 python tools/train.py ./configs/UFPMP-Det/coarse_det.py

Train MP-Det

CUDA_VISIBLE_DEVICES=0 python tools/train.py ./config/UFPMP-Det/mp_det_res50.py

Test

CUDA_VISIBLE_DEVICES=2 python UFPMP-Det-Tools/eval_script/ufpmp_det_eval.py \
    ./configs/UFPMP-Det/coarse_det.py \
    ./work_dirs/coarse_det/epoch_12.pth \
    ./configs/UFPMP-Det/mp_det_res50.py  \
    ./work_dirs/mp_det_res50/epoch_12.pth \
    XXXXX/dataset/COCO/annotations/instances_UAVval_v1.json \
    XXXXX/dataset/COCO/images/UAVval

Citation

If you find our paper or this project helps your research, please kindly consider citing our paper in your publication.

@inproceedings{ufpmpdet,
  title={UFPMP-Det: Toward Accurate and Efficient Object Detection on Drone Imagery},
  author={Huang, Yecheng and Chen, Jiaxin and Huang, Di},
  booktitle={AAAI Conference on Artificial Intelligence},
  year={2022}
}