README.md

metadata

dataset_info:
  - config_name: augmented
    features:
      - name: image
        dtype: image
      - name: objects
        struct:
          - name: bbox
            list:
              list: float64
          - name: categories
            list:
              class_label:
                names:
                  '0': trunk
                  '1': tiny_grape_bunch
                  '2': medium_grape_bunch
      - name: capture_set
        dtype: string
      - name: augmentation
        dtype: string
    splits:
      - name: train
        num_bytes: 2696503818
        num_examples: 18515
    download_size: 2694490286
    dataset_size: 2696503818
  - config_name: raw
    features:
      - name: image
        dtype: image
      - name: objects
        struct:
          - name: bbox
            list:
              list: float64
          - name: categories
            list:
              class_label:
                names:
                  '0': trunk
                  '1': tiny_grape_bunch
                  '2': medium_grape_bunch
      - name: capture_set
        dtype: string
    splits:
      - name: train
        num_bytes: 870691241
        num_examples: 2846
    download_size: 870301684
    dataset_size: 870691241
configs:
  - config_name: augmented
    data_files:
      - split: train
        path: augmented/train-*
  - config_name: raw
    data_files:
      - split: train
        path: raw/train-*
    default: true
license: cc-by-4.0
task_categories:
  - object-detection
size_categories:
  - 1K<n<10K

Vineyard Grape Trunk Detection

Images of vineyards in the Douro region of Portugal, taken by several ground-level cameras in varied illumination, with bounding boxes for vine trunks and for grape bunches at two growth stages (tiny, after bloom, and medium). The dataset accompanies a study on detecting grape bunches with quantized deep-learning models on low-power hardware; the current Zenodo version also adds trunk annotations. The raw config holds the 2,846 unaugmented RGB images; the augmented config holds 18,515 augmented copies (rotation, flipping, hue/saturation shifts, noise, scaling and translation), with the augmentation named per image. Each image records its capture set, and the record's thermal images are left out. The dataset contains raw and augmented versions. The raw split contains 2,846 images with 13,218 bounding box annotations across 3 categories.

The augmented split contains 18,515 images with 85,722 bounding box annotations across 3 categories.

This dataset is indexed on https://project-agml.github.io/ as part of the AgML python library.

Citation

@article{aguiar2021grape,
  title={Grape Bunch Detection at Different Growth Stages Using Deep Learning Quantized Models},
  author={Aguiar, André Silva and Magalhães, Sandro Augusto and dos Santos, Filipe Neves and Castro, Luis and Pinho, Tatiana and Valente, João and Martins, Rui and Boaventura-Cunha, José},
  journal={Agronomy},
  volume={11},
  pages={1890},
  year={2021},
  publisher={MDPI}
}

The dataset itself can be cited as:

Aguiar, A. S., & Magalhães, S. (2021). Grape bunch and vine trunk dataset for Deep Learning object detection. [Dataset]. Zenodo. https://doi.org/10.5281/ZENODO.5139598

This dataset was reformatted from its original format to match HuggingFace standards.