This repository includes the official implementation of the paper:

Dynamic Color Transform Networks for Wheat Head Detection

Plant Phenomics, 2022

& Dynamic Color Transform for Wheat Head Detection

International Conference on Computer Vision Workshops (ICCVW), 2021

Chengxin Liu, Kewei Wang, Hao Lu, Zhiguo Cao

Huazhong University of Science and Technology, China

• DCT is intuitive and steerable
• We won 2nd Place in Global Wheat Challenge 2021 based on DCT
• DCT is also applicable to other vision tasks, e.g., crowd counting

• Set up environment

# env
conda create -n dct python=3.7
conda activate dct

# install pytorch
conda install pytorch==1.10.0 torchvision==0.11.0 -c pytorch -c conda-forge

• Install

# clone 
git clone https://github.com/cxliu0/DCT.git
cd DCT

# install dependecies
pip install -r requirements/build.txt

• Download Global Wheat Head Detection Dataset 2021 (GWHD 2021). We expect the direcory structure to be as follow:

DCT
├── data
│    ├── gwhd_2021
│         ├── images
│         ├── competition_test.csv
│         ├── competition_train.csv
│         ├── competition_val.csv
├── configs
├── models
├── ...

• Preprocess data

python preprocess_data.py

• Download pretrained yolov4-p7 model and put it into ./pretrained/

We follow a two-step training stretegy.

• Step 1: train a baseline Scaled-YOLOv4 model

sh train_baseline.sh

• Step 2: fix the weights of the baseline model, and train DCT:

sh train_dct.sh

We implement the evaluation of ADA following Global Wheat Challenge 2021 (GWC 2021). Note that the tested ADA of this repository seems slightly lower than the official one.

• Modify test_ADA.sh

weights: modify it to the path of your locally trained model

Alternatively, pretrained DCT model is available at Google Drive.

• Run

sh test_ADA.sh

If you find this work or code useful for your research, please consider citing:

@article{liu2022dct,
  title = {Dynamic Color Transform Networks for Wheat Head Detection},
  author = {Chengxin Liu  and Kewei Wang  and Hao Lu  and Zhiguo Cao },
  journal = {Plant Phenomics},
  year={2022}
}

@INPROCEEDINGS{liu2021dct,
  author={Liu, Chengxin and Wang, Kewei and Lu, Hao and Cao, Zhiguo},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision Workshops (ICCVW)}, 
  title={Dynamic Color Transform for Wheat Head Detection}, 
  year={2021},
  pages={1278-1283},
}

This repository is based on Scaled-YOLOv4.