CitDet: A Benchmark Dataset for Citrus Fruit Detection

Overview

Despite the fact that significant progress has been made in solving the fruit detection problem, the lack of publicly available datasets has complicated direct comparison of results. For instance, citrus detection has long been of interest to the agricultural research community, yet there is an absence of work, particularly involving public datasets of citrus affected by Huanglongbing (HLB).

This repository provides source code for our paper titled "CitDet: A Benchmark Dataset for Citrus Fruit Detection." The CitDet dataset contains high-resolution images of citrus trees located in an area known to be highly affected by HLB, along with high-quality bounding box annotations of citrus fruit. Fruit on both the trees and the ground are labeled to allow for identification of fruit location, which contributes to advancements in yield estimation and potential measure of HLB impact via fruit drop. In addition, we enhance state-of-the-art object detection methods for use in typical orchard settings and provide new benchmark results.

More information on the project can be found on the CitDet website.

Citation

If you find this project useful, then please consider citing both our paper and dataset.

@article{james2024citdet,
  title={CitDet: A Benchmark Dataset for Citrus Fruit Detection},
  author={James, Jordan A and Manching, Heather K and Mattia, Matthew R and Bowman, Kim D and Hulse-Kemp, Amanda M and Beksi, William J},
  journal={arXiv preprint arXiv:2309.05645},
  year={2024}
}

@data{T8/QFVHQ5_2024,
  title={{CitDet}},
  author={James, Jordan A and Manching, Heather K and Mattia, Matthew R and Bowman, Kim D and Hulse-Kemp, Amanda M and Beksi, William J},
  publisher={Texas Data Repository},
  version={V1},
  url={https://doi.org/10.18738/T8/QFVHQ5},
  doi={10.18738/T8/QFVHQ5},
  year={2024}
}

Installation

First, begin by cloning the project:

$ git clone https://github.com/robotic-vision-lab/CitDet-A-Benchmark-Dataset-For-Citrus-Fruit-Detection.git
$ cd CitDet-A-Benchmark-Dataset-For-Citrus-Fruit-Detection

Then, set up a Python 3 environment and install Pytorch (1.0.1 or higher) and TorchVision. Finally, install the remaining packages:

$ pip install Pillow opencv-python scikit-learn numpy

Dataset

The dataset can be downloaded from the Texas Data Repository or the [USDA Ag Data Commons.] (https://doi.org/10.15482/USDA.ADC/1529611) The dataset contains images in JPG format, bounding box annotations in COCO JSON format, and pseudo-labels for semantic segmentation, generated by SAM, in PNG format.

Data Loader

The script datasets/coco.py contains a COCO dataset class that allows for loading images and masks on the fly, and extracting bounding boxes and segmentation masks. Modify this class accordingly if you need additional inputs to your network.

Training, Evaluation, and Visualization

To train a network on CitDet, first download the dataset. The notebook train_eval_visualize.py includes example code for training, evaluating, and visualizing the predictions of a detection model. Additionally, we provide an evaluation script, eval_coco_json.py.

The following command can be used to run the evaluation script:

$ python eval_coco_json.py --pred_file path/to/preds.json --gt_file path/to/ground_truth.json

Evaluation

Here are our latest models along with their respective benchmarks.

Whole Image Dataset Results

Method	Backbone	AP	AP @ IoU=.50	AP @ IoU=.75
FasterRCNN	ResNet50	0.220	0.515	0.140
YOLOv5-m	YOLOv5	0.348	0.700	0.298
YOLOv7-m	YOLOv7	0.406	0.779	0.367

Tiled Image Dataset Results

Method	Backbone	AP	AP @ IoU=.50	AP @ IoU=.75
YOLOS	ViT	0.324	0.707	0.246
DETR	ResNet50	0.350	0.728	0.288
FasterRCNN	ResNet50	0.372	0.760	0.315
YOLOv5-m	YOLOv5	0.449	0.819	0.434
YOLOv7-m	YOLOv7	0.455	0.831	0.439

CitDet Source Code License

CitDet Dataset License