leofansq/fii-centernet

[IEEE TVT] FII-CenterNet: an anchor-free detector with foreground attention for traffic object detection

FII-CenterNet: An Anchor-free Detector with Foreground Attention for Traffic Object Detection

For technical details, please refer to:

FII-CenterNet: An Anchor-free Detector with Foreground Attention for Traffic Object Detection

(0) Abstract

Most successful object detectors are anchor-based, which is difficult to adapt to the diversity of traffic objects. We propose a novel anchor-free method, called FIICenterNet. It introduces the foreground information to CenterNet to eliminate the interference of complex background information in traffic scenes. The foreground region proposal network is based on semantic segmentation. Midground is introduced as the transition between foreground and background. In addition to foreground location information, scale information is introduced to improve the regression results.

(1) Setup

This code has been tested with Python 3.6, PyTorch v0.4.1, CUDA 9.0 and cuDNN v7 on Ubuntu 16.04.

• Setup python environment

  conda create -n FIICenterNet python=3.6
  source activate FIICenterNet
  

• Install pytorch0.4.1

  conda install pytorch=0.4.1 torchvision -c pytorch
  

• Install COCOAPI

  # COCOAPI=/path/to/clone/cocoapi
  git clone https://github.com/cocodataset/cocoapi.git $COCOAPI
  cd $COCOAPI/PythonAPI
  make
  python setup.py install --user
  

• Clone the repository

• Install the requirements

  pip install -r requirements.txt
  

• Compile deformable convolutional (from DCNv2)

  cd $FIICenterNet_ROOT/src/lib/models/networks/DCNv2
  ./make.sh
  

• [Optional] Compile NMS if your want to use multi-scale testing or test ExtremeNe.

  cd $FIICenterNet_ROOT/src/lib/external
  make
  

• Download pertained models

  Dataset	Pretrained Model	Results
  KITTI	Baidu: 3xdh	Baidu: sb67
  Pascal VOC	Baidu: xhta	Baidu: f0oi

(2) KITTI

• Preparing the dataset

  Dataset is transformed to COCO foramt to train for now.

  • Download the dataset KITTI
  • File structure

    FII-CenterNet Root
    |---- data
         |---- kitti
               |---- training
                     |---- Labels (containing download labels txt)
         |---- KITTI_COCO
               |---- annotations (containing transformed jason)
               |---- images (containing download taining images)
               |---- test_images (conmtaining download testing images)
    

    transformed annotations Baidu: t55q

• Training

  python main.py ctdet --exp_id kitti_exp --dataset=kitti --batch_size 8 --gpus 0 --attention --wh_weight=0.1 --lr=1e-4
  

  For Evaluation: rename ./src/lib/datasets/dataset/kitti_half.py to kitti.py

  For Testing: rename ./src/lib/datasets/dataset/kitti_full.py to kitti.py

• Evaluation

  • Predict

  python test.py --exp_id kitti_exp --not_prefetch_test ctdet --dataset=kitti --load_model /your_path/your_model.pth --attention --flip_test
  

  • Evaluate

  ./tools/kitti_eval/evaluate_object_3d_offline ../data/kitti/training/Labels/ /your_path/results/
  

• Testing

  python test.py --exp_id kitti_full --not_prefetch_test ctdet --dataset=kitti --load_model /your_path/model_last.pth --attention --flip_test --trainval
  

(3) Pascal VOC

• Preparing the dataset

  Train on VOC 2007 and VOC 2012 trainval sets and test on VOC 2007 test set. Dataset is transformed to COCO foramt to train for now.

  • Download the dataset
  • File structure

    FII-CenterNet Root
    |---- data
         |---- VOC_COCO
               |---- Annotations (containing download annotations xml)
               |---- annotations (containing transformed jason)
               |---- images (containing download taining images)
    

    transformed annotations Baidu: qbhp

• Training

  python main.py ctdet --exp_id voc_exp --batch_size 32 --gpus 0,1 --attention --wh_weight=0.1 --lr=1.25e-4 --num_epochs=140 --lr_step='90,120'
  

• Evaluation

  • Predict

  python test.py --exp_id voc_exp --not_prefetch_test ctdet --load_model /your_path/your_model.pth --attention --flip_test --trainval
  

  • Evaluate

  python tools/reval.py /your_path/results.json
  

(4) Visualization

python demo.py ctdet --demo /.../FII-CenterNet/kitti_images/ --load_model /your_path/your_model.pth --attention --debug=2

Citation

If you find our work useful in your research, please consider citing:

@ARTICLE{9316984,
  author={Fan, Siqi and Zhu, Fenghua and Chen, Shichao and Zhang, Hui and Tian, Bin and Lv, Yisheng and Wang, Fei-Yue},
  journal={IEEE Transactions on Vehicular Technology}, 
  title={FII-CenterNet: An Anchor-Free Detector With Foreground Attention for Traffic Object Detection}, 
  year={2021},
  volume={70},
  number={1},
  pages={121-132},
  doi={10.1109/TVT.2021.3049805}}

Acknowledgment

Part of our code refers to the recent work Objects as Points.

fii-centernet's Issues

Importerror: undefinded symbol

您好，

我对您发表的文章和网络模型十分感兴趣，并且我按照您Readme中提供的指令配置了环境，但我在进行测试和评估kitti的环节遇到了这样的问题，想问下要如何解决呢？ 谢谢！