This project contains the code for the paper "Shooting Condition Insensitive Unmanned Aerial Vehicle Object Detection" and relies on mmdetection 2.x.
We recommend configuring mmdetection according to https://github.com/open-mmlab/mmdetection/tree/2.x, and for other environment and dependency setups, please refer to the environment specified in requirements.txt.
After preparing the environment, you can download the following datasets:
You can also convert the dataset annotations to the COCO format. For your convenience, we have provided COCO format annotations in the project.
Please organize the datasets as follows:
DATA
├─ UAVDT
│ ├─ annotations
│ │ ├─ UAVDT_test_coco.json
│ │ ├─ UAVDT_train_coco.json
│ ├─ images
│ │ ├─ test
│ │ │ ├─ M0203
│ │ │ │ ├─ xxxx.jpg
│ │ │ │ └─ ...
│ │ │ └─ ...
│ │ ├─ train
│ │ │ ├─ M0101
│ │ │ │ ├─ xxxx.jpg
│ │ │ │ └─ ...
│ │ │ └─ ...
├─ visdrone_coco
│ ├─ annotations
│ │ ├─ instances_UAVval.json
│ │ ├─ instances_UAVtrain.json
│ ├─ images
│ │ ├─ instances_UAVtrain
│ │ │ ├─ xxxx.jpg
│ │ │ └─ ...
│ │ ├─ instances_UAVval
│ │ │ ├─ xxxx.jpg
│ │ │ └─ ...- Run
text_learner/gen_fix_prompts.pyto generate initial prompt features. - Then, run
text_learner/prompts_learner.pyto generate fine-tuned features.
Please note that training is supported on a single GPU:
python train.py configs/xxxx.pyIf you find this repository helpful, please consider citing our paper:
@article{LIU2024123221,
title = {Shooting Condition Insensitive Unmanned Aerial Vehicle Object Detection},
journal = {Expert Systems with Applications},
volume = {246},
pages = {123221},
year = {2024},
issn = {0957-4174},
doi = {https://doi.org/10.1016/j.eswa.2024.123221},
url = {https://www.sciencedirect.com/science/article/pii/S0957417424000861},
author = {Jie Liu and Jinzong Cui and Mao Ye and Xiatian Zhu and Song Tang},
}
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent