⭐ This code has been completely released ⭐

⭐ our article ⭐

If our code is helpful to you, please cite:

@article{zhang2022superyolo,
  title={SuperYOLO: Super Resolution Assisted Object Detection in Multimodal Remote Sensing Imagery},
  author={Zhang, Jiaqing and Lei, Jie and Xie, Weiying and Fang, Zhenman and Li, Yunsong and Du, Qian},
  journal={arXiv preprint arXiv:2209.13351},
  year={2022}
}

pip install -r requirements.txt

• 1.1 In order to realize the SR assisted branch, the input images of the network are downsampled from 1024 x 1024 size to 512 x 512 during the training process. In the test process, the image size is 512 x 512, which is consistent with the input of other algorithms compared.

• 1.2 Download VEDAI data for our experiment from baiduyun (code: hvi4) or google drive. And the path of dataset is like that

SuperYOLO
├── dataset
│   ├── VEDAI
│   │   ├── images
│   │   ├── labels
│   │   ├── fold01.txt
│   │   ├── fold01test.txt
│   │   ├── fold02.txt
│   │   ├── .....
│   ├── VEDAI1024
│   │   ├── images
│   │   ├── labels

• 1.3 Note that we transform the labels of the dataset to be horizontal boxes by transform code. You shoud run transform.py before training the model. Change the PATH = './dataset/' and then run the code.

python train.py --cfg models/SRyolo_noFocus_small.yaml --super --train_img_size 1024 --hr_input --data data/SRvedai.yaml --ch 4 --input_mode RGB+IR

new fusion method MF

python train.py --cfg models/SRyolo_MF.yaml --super --train_img_size 1024 --hr_input --data data/SRvedai.yaml --ch 64 --input_mode RGB+IR+MF

You can use our pretrained checkpoints for test process. Download pre-trained model and put it in here.

!!!⭐⭐ The new weight is available at small_EDSR_MF

The google drive link is small_EDSR_MF

python test.py --weights runs/train/exp/best.pt --input_mode RGB+IR 

new fusion method MF

python test.py --weights runs/train/exp/best.pt --input_mode RGB+IR+MF

2023.2.14 open the train.py

2023.2.14 update the new fusion method (MF)

2023.2.16 update the test.py for visualization of detection results

This code is built on YOLOv5 (PyTorch). We thank the authors for sharing the codes.

If you have any question, please contact me with email ([email protected]).