Wheat Head Detection

This page is resource for AICrowd: Global Wheat Challenge 2021

Global Wheat Head Detection (GWHD) Dataset: A Large and Diverse Dataset of High-Resolution RGB-Labelled Images to Develop and Benchmark Wheat Head Detection Methods

Previous Competition Solution

Kaggle Global Wheat Detection 2020

1st place solution

https://www.kaggle.com/c/global-wheat-detection/discussion/172418

Summary

• Custom mosaic data augmentation
• MixUp
• Heavy augmentation
• Data cleaning
• EfficientDet
• Faster RCNN FPN
• Ensemble multi-scale model: Weighted-Boxes-Fusion, special thank @zfturbo
• Test time augmentation(HorizontalFlip, VerticalFlip, Rotate90)
• Pseudo labeling

2nd Place Solution

Link: https://www.kaggle.com/c/global-wheat-detection/discussion/175961

Repository: https://github.com/liaopeiyuan/TransferDet

3rd Place Solution

Code repository https://github.com/ufownl/global-wheat-detection

• YOLOv3 from GluonCV
• Use Darknet53 backbone
• Use WBF over TTA
• Use pseudo-labeling technique

Papers & Literature

• EfficientDet: Scalable and Efficient Object Detection
• An analysis of last year competition: [2105.06182] Global Wheat Challenge 2020: Analysis of the competition design and winning models (arxiv.org)
• An introduction on the new datasets : [2105.07660] Global Wheat Head Dataset 2021: an update to improve the benchmarking wheat head localization with more diversity (arxiv.org)
• Roman Solovyev, Wimin Wang, Tatiana Gbruseva, Weighted boxes fusion: Ensembling boxes from different object detection models with [Code]
• Global Wheat Head Detection (GWHD) dataset: a large and diverse dataset of high resolution RGB labelled images to develop and benchmark wheat head detection methods https://hal.archives-ouvertes.fr/hal-02560389
• Wheat Head Detection using Deep, Semi-Supervised and Ensemble Learning https://www.tandfonline.com/doi/abs/10.1080/07038992.2021.1906213?journalCode=ujrs20
• Bo Gong, Daji Ergu, Ying Cai, Bo Ma, Real-Time Detection for Wheat Head Applying Deep Neural Network https://pubmed.ncbi.nlm.nih.gov/33396711/
• Saeed Khaki, Nima Safaei, Hieu Pham, Lizhi Wang, WheatNet: A Lightweight Convolutional Neural Network for High-throughput Image-based Wheat Head Detection and Counting https://arxiv.org/abs/2103.09408
• Md Mehedi Hasan, Joshua P. Chopin, Hamid Laga & Stanley J. Miklavcic , Detection and analysis of wheat spikes using Convolutional Neural Networks https://plantmethods.biomedcentral.com/articles/10.1186/s13007-018-0366-8
• Do Thuan, EVOLUTION OF YOLO ALGORITHM AND YOLOV5: THE STATE-OF-THE-ART OBJECT DETECTION ALGORITHM , https://www.theseus.fi/bitstream/handle/10024/452552/Do_Thuan.pdf

Code

• EfficientDet (Scalable and Efficient Object Detection) implementation in Keras and Tensorflow
• Signatrix/efficientdet: Pytorch implementation
• xuannianz/EfficienDet: Implementation on Keras & Tensorflow
• Google: EfficientDet
• rwightman/efficientdet : A PyTorch implementation of EfficientDet. (used in 1st & 2nd place Kaggle 2020)
• YoloV5 https://github.com/ultralytics/yolov5
• Global Wheat Competition 2021, Starting Notebook using the WILDS library

Yolo5 was not eligible for 2020 Wheat Head Challenge, but it can be used in 2021 Wheat Head Challenge [discussion]

Articles

• EfficientDet: Towards Scalable and Efficient Object Detection
• A Throrough Breakdown of EfficientDet for Object Detection

Wheat Dataset

• Global Wheat Dataset http://www.global-wheat.com/
• Global Wheat Head Dataset – 2020 challenge version
• [Paper: Detection and Analysis of Wheat Spikes Using Convolutional Neural Networks], [Dataset Link]
• Wheat Dataset https://plantimages.nottingham.ac.uk/