Home

Awesome

TricubeNet - Official Pytorch Implementation (WACV 2022)

TricubeNet: 2D Kernel-Based Object Representation for Weakly-Occluded Oriented Object Detection <br /> Beomyoung Kim<sup>1</sup>, Janghyeon Lee<sup>2</sup>, Sihaeng Lee<sup>2</sup>, Doyeon Kim<sup>3</sup>, Junmo Kim<sup>3</sup><br>

<sup>1</sup> <sub>NAVER CLOVA</sub><br /> <sup>2</sup> <sub>LG AI Research</sub><br /> <sup>3</sup> <sub>KAIST</sub><br />

WACV 2022 <br />

Paper

PWC PWC

<img src = "https://github.com/qjadud1994/TricubeNet/blob/main/figures/overview.png" width="100%" height="100%">

How to use?

Data Preparation

For training

bash run_train.sh

Please check the discription of training hyperparameters (we recommend to use default hyperparameters)

python3 train.py --help

For testing

cd evaluation
bash run_eval.sh

Please check the discription of testing hyperparameters (we recommend to use default hyperparameters)

python3 eval_DOTA.py --help

Qualitative Results

DOTA

<img src = "https://github.com/qjadud1994/TricubeNet/blob/main/figures/DOTA.png" width="70%" height="70%">

MSRA-TD500, ICDAR 2015

<img src = "https://github.com/qjadud1994/TricubeNet/blob/main/figures/Text-Detection.png" width="70%" height="70%">

SKU110K-R

<img src = "https://github.com/qjadud1994/TricubeNet/blob/main/figures/SKU110K-R.png" width="70%" height="70%">

Citation

We hope that you find this work useful. If you would like to acknowledge us, please, use the following citation:

@inproceedings{kim2022tricubenet,
  title={TricubeNet: 2D Kernel-Based Object Representation for Weakly-Occluded Oriented Object Detection},
  author={Kim, Beomyoung and Lee, Janghyeon and Lee, Sihaeng and Kim, Doyeon and Kim, Junmo},
  booktitle={Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision},
  pages={167--176},
  year={2022}
}