AgrilPlant dataset

Download:

AgrilPlant dataset (train/test)
AgrilPlant dataset (5-folds)

Description:

The AgriPlant dataset consists of 3,000 agriculture images that are collected from the website www.flickr.com. It consists of 10 classes with the following plants: apple, banana, grape, jackfruit, orange, papaya, persimmon, pineapple, sunflower, and tulip. Each class contains exactly 300 images. The images may have been taken from five different views, i.e. entire plant, branch, flower, fruit, and leaf.

The AgrilPlant dataset faces some challenges due to the following reasons:
1) a dissimilarity of plants within the same class, for example, there are varieties of shape and color of tulips, or there are several colors of apples,
2) a similarity among some classes, for example, apple, orange, and persimmon images consist of similar shapes and colors, and
3) the complex backgrounds in most of the images.

The dataset is splitted with a ratio of 70:30 for train and testing set.

Publications:

Citation:

@inproceedings{pawara2017comparing,
  title={Comparing Local Descriptors and Bags of Visual Words to Deep Convolutional Neural Networks for Plant Recognition.},
  author={Pawara, Pornntiwa and Okafor, Emmanuel and Surinta, Olarik and Schomaker, Lambert and Wiering, Marco},
  booktitle={ICPRAM},
  pages={479--486},
  year={2017}
}

Tropic dataset

Download:

dataset_tropic20 (train/test)
dataset_tropic10 (5-folds)
dataset_tropic52: exp1
dataset_tropic52: exp2
dataset_tropic52: exp3
dataset_tropic52: exp4
dataset_tropic52: exp5
source code


Description:

The Tropic dataset contains 20 classes of plants with a total of 5,276 images. Each of the class contain non-uniform distribution of images which varies from 221 to 371 images per class. The dataset contains the following plants:

  • Acacia
  • Ashoka
  • Bamboo
  • Banyan
  • Chinese wormwood
  • Croton
  • Crown flower
  • Ervatamia
  • Golden shower
  • Hibiscus
  • Lady palm
  • Lime
  • Mango
  • Manila tamarind
  • Poinsettia
  • Raspberry ice Bougainvillea
  • Sanchezia
  • Umbrella tree
  • West Indian jasmine
  • White plumeria
The images were collected during the day using the DSLR camera. The data collection was done from diverse locations in Northeastern Thailand. All the images have similarities in illumination conditions but dissimilar plant parts (flowers, branches, fruits, leaves, or the whole tree) and background information such as sky, houses, and soil. We randomly split the dataset in the ratio 70:30 for training and testing set, respectively.

Publications:

  • Pawara, P.,Okafor, E., Groefsema, M., He, S., Schomaker, L, Wiering, M. (2020). One-vs-One Classification for Deep Neural Networks.Pattern Recognitionใ

Citation:

@article{pawara2020one,
  title={One-vs-One Classification for Deep Neural Networks},
  author={Pawara, Pornntiwa and Okafor, Emmanuel and Groefsema, Marc and He, Sheng and Schomaker, Lambert RB and Wiering, Marco A},
  journal={Pattern Recognition},
  pages={107528},
  year={2020},
  publisher={Elsevier}
}

AgrilFruit dataset - for object detection and counting task

Download:

Click here to download the AgrilFruit dataset
AgrilFruit-augmented exp1
AgrilFruit-augmented exp2
AgrilFruit-augmented exp3
AgrilFruit-augmented exp4
AgrilFruit-augmented exp5
source code


Description:

The AgrilFruit dataset contains 5 classes of fruit images. Each class consists of 300 images, which 240 images are used for training and the rest 60 images are used for testing. The dataset is used mainly for the object detection and fruit counting tasks. So, I also provide the xml files and the text files.

Publications:

Citation:

To be published on October 2020.

Monkey10 and uMonkey10 - finegrained dataset

Download:

Monkey10 (train/test)
Monkey10 (5-folds) 2.9 GB
uMonkey10 (Unbalanced Monkey)(train/test)
uMonkey10 (5-folds) 1.6 GB

Description:

The Monkey-10 dataset is orginally from kaggle. It contains approximately 1,400 images and 10 classes, and each class corresponds to a different species of monkeys. Each of the classes contains approximately 110 training images and 27 test images. The dataset consists of the following monkey species:

  • Mantled howler
  • Patas monkey
  • Bald uakari
  • Japanese macaque
  • Pygmy marmoset
  • White-headed capuchin
  • Silvery marmoset
  • Common squirrel monkey
  • Black-headed night monkey
  • Nilgiri langur
The Monkey-10 dataset was primarily used to observe if performance differences between the OvO and OvA schemes generalize to a different kind of fine-grained species dataset.
Additionally, from the original Monkey-10 dataset, we randomly selected a non-uniform distribution of images from the training set, which varies from 10 to 120 images per class to create an imbalanced dataset. This dataset is called Imbalanced-Monkey-10 and serves as a purpose to study if the OvO or OvA scheme can better handle strongly imbalanced classes.

Publications:

@article{pawara2020one,
  title={One-vs-One Classification for Deep Neural Networks},
  author={Pawara, Pornntiwa and Okafor, Emmanuel and Groefsema, Marc and He, Sheng and Schomaker, Lambert RB and Wiering, Marco A},
  journal={Pattern Recognition},
  pages={107528},
  year={2020},
  publisher={Elsevier}
}