Weakly Supervised Rotation-Invariant Aerial Object Detection Network

By Xiaoxu Feng, Xiwen Yao, Gong Cheng, Junwei Han

We have released the codes of IENet work here. It is the extension of RINet and obtains state-of-the-art performance on the PASCAL VOC and MS COCO!

Citation

@InProceedings{Feng_2022_CVPR,
    author    = {Feng, Xiaoxu and Yao, Xiwen and Cheng, Gong and Han, Junwei},
    title     = {Weakly Supervised Rotation-Invariant Aerial Object Detection Network},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2022},
    pages     = {14146-14155}
}

Overview

The code will be released soon.

Requirements

• python == 3.6
  
• Cuda == 9.0
  
• Pytorch == 0.4.1
  
• torchvision == 0.2.1
  
• Pillow
  
• sklearn
  
• opencv
  
• scipy
  
• cython
  
• GPU: GeForce RTX 2080Ti | Tesla V100

  Installation

  1. Clone the RINet repository

     git clone https://github.com/XiaoxFeng/RINet.git

  2. Compile

     cd RINet/lib
     bash make.sh

     3.Download the VOCdevkit and rename it as VOCdevkit2007

     cd RINet/data/
     wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCdevkit_08-Jun-2007.tar

  3. Download the training, validation, test data from NWPU, NWPU.V2 and DIOR
  4. Extract all of datasets into one directory named VOCdevkit2007
  5. Download pretrained ImageNet weights from here, and put it in the data/imagenet_weights/
  6. Download selective search proposals from NWPU and DIOR, and put it in the data/selective_search_data/

     Train model

     ./experiments/scripts/train_faster_rcnn.sh 0 pascal_voc vgg16

     Test model

     ./experiments/scripts/test_faster_rcnn.sh 0 pascal_voc vgg16

     Download models

     Models trained on DIOR can be downloaded here:Google Drive.

     Acknowledgement

     We borrowed code from MLEM, PCL, and Faster-RCNN.