- News: The preprint version of our paper is now available here.
We present the Deep Nutrient Deficiency - Dikopshof - Winter Wheat and Winter Rye (DND-Diko-WWWR) dataset, which consists of 1,800 RGB images of winter wheat (WW2020) and 1,800 RGB images of winter rye (WR2021). The images were captured by a camera mounted on a UAV at three time points at the long-term fertilizer experiment (LTFE) Dikopshof near Bonn, Germany. The images were annotated with seven types of fertilizer treatments. The dataset is used for image classification.
Example images
The dataset can be downloaded from PhenoRoam.
The dataset directory should have this basic structure:
DND-Diko-WWWR/
├── WW2020/
├── images/
├── 20200422_0.jpg
└── ...
├── trainval.txt
├── test.txt
└── labels_trainval.yml
└── WR2021/
...
- UAV-Based RGB Images: The proposed DND-Diko-WWWR dataset consists of 1,800 RGB images of winter wheat (WW2020) and 1,800 RGB images of winter rye (WR2021).
- Annotation: The images were annotated with seven nutrient treatments: unfertilized, PKCa, N_KCa, NP_Ca, NPK, NPKCa, NPKCa+m+s, where “_” stands for the omission of the corresponding nutrient (N: nitrogen, P: phosphorous, K: potassium, Ca: lime) and +m+s stands for additional application of mineral fertilizer and farmyard manure.
-
File Name: The file was named as Date_RandomNumber.jpg, e.g. 20200506_0.jpg, where Date is the sampling date and RandomNumber
$r \in[0, 1800)$ . - #Train:#Val:#Test: 1332:468 (74%:26%).
As part of the 8th Workshop on Computer Vision in Plant Phenotyping and Agriculture (CVPPA) at the IEEE/CVF International Conference of Computer Vision (ICCV) 2023, we are organizing a challenge that aims at providing a solution to recognize nutrient deficiencies in winter wheat winter rye using UAV-based RGB images.
In this challenge, one has to provide a zipped file of two text file named "predictions_WW2020.txt" and "predictions_WR2021.txt" with no headers. Each line in the text file should state the image filename, followed by the prediction index (0-6). The file should contain a prediction for each of the test images, corresponding to 468 lines of text. A sample submission file is shown below:
submission.zip
- predictions_WW2020.txt
20200506_0.jpg 5
20200506_11.jpg 0
...
20200422_1797.jpg 6
- predictions_WR2021.txt
An example of submission file can be download here.
Teams will compete to provide the highest top-1 accuracy on the subset of winter wheat (WW2020) and winter rye (WR2021), respectively. The ranking will be based on the average accuracy of WW2020 and WR2021.
The best-performing solutions will be contacted by the workshop organizers after the end date of the challenge and we will have a part in the workshop to announce the winners of this challenge. Besides a certificate, the authors are invited to provide an overview of their technical approach (and codes with reproducible results).
Note
- It is allowed to use publicly available pre-trained models for weight initialization and transfer learning purposes.
- However, it is NOT permitted to use other data than the provided dataset for this competition.
- It is the participant's responsibility to divide the training set into proper training and validation splits if it is needed
Please specify your data_root (path to the downloaded dataset) at ./configs/config.py. We provide a baseline model based on ResNet-50 and Swin Transformer V2. You can now use the script trainval.sh to train the baseline model, and test.sh to perform prediction on the test set and to generate the .txt file for submission to Codalab.
The top-1 accuracy is used. We provide an evaluation script to test your model over the validation set (a subset of the provided trainval set generated by yourself). Note that this script cannot be used to evaluate models over the test set, as we do not provide labels for the test set. It is good practice to ensure your predictions work with this script, as the same script is used on the evaluation server.
Run python ./codalab/evaluate.py with the arguments as described in the script to evaulate the predictions. This creates a scores.txt file containing the metrics used for the challenge leaderboard.
The evaluation server is hosted using CodaLab. Submitting to the challenge requires a CodaLab account. Please find the evaluation server here. To participate in the challenge, you need to upload a .zip file of the .text files generated by the script provided before.
Q: Can we use any other data than the data provdided? A: No.
Q: Is data augmentation allowed? A: Yes, as long as the augmentations are only applied to the provided data.
Q: Can we use two different models for WW2020 and WR2021? A: No, the models should have the same architecture, but you can finetune both models separately.
If you find this dataset useful for your research, please cite our paper:
@article{yi4549653non,
title={Non-Invasive Diagnosis of Nutrient Deficiencies in Winter Wheat and Winter Rye Using Uav-Based Rgb Images},
author={Yi, Jinhui and Lopez, Gina and Hadir, Sofia and Weyler, Jan and Klingbeil, Lasse and Deichmann, Marion and Gall, Juergen and Seidel, Sabine J},
journal={Available at SSRN 4549653}
}
@article{yi2020deep,
title={Deep Learning for Non-Invasive Diagnosis of Nutrient Deficiencies in Sugar Beet Using RGB Images},
author={Yi, Jinhui and Krusenbaum, Lukas and Unger, Paula and Hüging, Hubert and Seidel, Sabine J and Schaaf, Gabriel and Gall, Juergen},
journal={Sensors},
volume={20},
pages={5893},
year={2020},
}
This dataset follows Creative Commons Attribution Non Commercial Share Alike 4.0 Internation License.