Video-To-Dataset-Orchard-Apple Dataset

Dataset Description

This dataset provides a collection of representative image frames extracted from drone video sequences captured in various agricultural environments, with a particular focus on apple orchards. It also includes scenes from grasslands and panoramic views. The frame extraction and selection process followed the "Video-To-Dataset" methodology developed by the primary author.

The dataset is designed to support research and development in agricultural computer vision, including tasks like:

• Fruit detection and counting (apples)
• Crop monitoring and yield estimation
• Plant phenotyping
• Weed and disease detection
• Testing and benchmarking video analysis algorithms for agricultural applications

This dataset combines personally collected data by the dataset curator with publicly available data sourced from the USDA Ag Data Commons. Users must be aware of the different licensing terms applicable to each part of the dataset (see Licensing Information).

Dataset Curator: Miroslav Jaros (Mendel University in Brno)

Data Sources

The data originates from two sources:

1. Personally Collected Data (by Miroslav Jaros)

• Description: Drone video footage was captured by Miroslav Jaros at various agricultural sites, primarily in the Czech Republic. This source corresponds to the following scene folders:
  • liskovec_apple_orchard
  • panorama
  • orchard_march
  • orchard_october_beginning
  • grasslands
• License: This portion of the data is made available under the Creative Commons Attribution 4.0 International License (CC BY 4.0). You are free to share and adapt this data for any purpose, provided you give appropriate credit to Miroslav Jaros and link to the license. See: https://creativecommons.org/licenses/by/4.0/

2. USDA Public Data

• Description: Video footage was obtained from a public dataset made available by the United States Department of Agriculture (USDA) Ag Data Commons. This source corresponds to the following scene folders:
  • usa_without_blue
  • usa_with_blue
• Original Source: Parton, J., Steward, B., Schumacher, L., & Dowell, F. (2018). Multispectral Videography Acquired from UAV during Corn and Soybean Growth. Ag Data Commons.
• DOI: https://doi.org/10.15482/USDA.ADC/1416008
• License: Crucially, the use of the data within the usa_without_blue and usa_with_blue folders is governed by the terms and conditions specified by the USDA at the original source DOI link. Users must consult and comply with the license and usage restrictions provided by the USDA for this specific data. The CC BY 4.0 license does not apply to this portion of the dataset.

Methodology

The image frames were extracted and selected using the pipeline detailed in the following publication:

Jaros, M., Slany, V., Skrob, O., Trenz, O., Stastny, J., Samarek, R., & Martinek, R. (2025). Methodology for Extracting Representative Frames from Video Sequences for Dataset Creation in Agricultural Applications. [JOURNAL/CONFERENCE NAME, VOLUME, PAGES]

Pipeline Summary:

1. Frame Extraction: Sampling frames from the source MP4 video.
2. Sharp Frame Detection: Filtering blurry frames using Gaussian-Laplacian variance.
3. Clustering: Vectorizing sharp frames (e.g., Lowres, ResNet18, CLIP) and clustering them using k-means with dynamically determined k (e.g., Silhouette score).
4. Representative Selection: Selecting the sharpest frame (Laplacian variance) from each cluster.

Dataset Structure

The dataset is organized by scene. Each scene folder contains subfolders with the extracted PNG frames (corresponding to different extraction parameter variants) and the original source MP4 video file.

The structure in the repository root is:

/
├── gauss_laplac-450/
│   ├── grasslands/
│   │   ├── final_dyn_clip/
│   │   │   └── frame_xxxx.png ...
│   │   ├── final_dyn_lowres/ ...
│   │   ├── final_dyn_rn18/ ...
│   │   ├── final_stat30_lowres/ ...
│   │   ├── final_stat60_clip/ ...
│   │   ├── final_stat60_rn18/ ...
│   │   └── xxxxx.MP4           # <<== Original MP4 video for this scene
│   ├── liskovec_apple_orchard/
│   │   ├── final_*/ ...        # (frame subfolders)
│   │   └── xxxxx.MP4           # <<== Original MP4 video
│   ├── orchard_march/
│   │   ├── final_*/ ...
│   │   └── xxxxx.MP4
│   ├── orchard_october_beginning/
│   │   ├── final_*/ ...
│   │   └── xxxxx.MP4
│   ├── panorama/
│   │   ├── final_*/ ...
│   │   └── xxxxx.MP4
│   ├── usa_without_blue/       # Derived from USDA data
│   │   ├── final_*/ ...
│   │   └── xxxxx.MP4           # <<== Original MP4 video (subject to USDA terms)
│   └── usa_with_blue/          # Derived from USDA data
│       ├── final_*/ ...
│       └── xxxxx.MP4           # <<== Original MP4 video (subject to USDA terms)
├── .gitattributes              # Handles Git LFS tracking
├── LICENSE                     # License file
└── README.md                   # This file (Dataset Card)

(Note: xxxxx.MP4 represents the specific source video file located in each scene directory. final_*/ represents the various subfolders containing extracted frames.)

Data Format

• Extracted Frames: PNG format (.png). Located within subfolders (e.g., final_dyn_clip) inside each main scene folder.
• Source Videos: MP4 format (.MP4). Located directly inside each main scene folder (e.g., grasslands/DJI_0278_W-008.MP4).

Licensing Information

This dataset contains data under two different licenses:

1. Personally Collected Data: The content within the following folders is licensed under CC BY 4.0:

   • liskovec_apple_orchard
   • panorama
   • orchard_march
   • orchard_october_beginning
   • grasslands (See: https://creativecommons.org/licenses/by/4.0/)

2. USDA Sourced Data: The content within the following folders is derived from data provided by the USDA Ag Data Commons and is subject to their original licensing terms:

   • usa_without_blue
   • usa_with_blue Users must refer to the original source for licensing details and comply accordingly: https://doi.org/10.15482/USDA.ADC/1416008. The CC BY 4.0 license does not apply to these specific folders.

By using this dataset, you agree to comply with the respective licenses for each portion of the data.

Citation Information

If you use this dataset or the methodology in your research, please cite the following:

1. This Dataset:

@misc{jaros_video_to_dataset_orchard_apple_2025,
  author       = {Jaros, Miroslav},
  title        = {Video-To-Dataset-Orchard-Apple},
  year         = {2025},
  publisher    = {Hugging Face},
  journal      = {Hugging Face Hub},
  howpublished = {\url{https://huggingface.co/datasets/miroslavjaros/Video-To-Dataset-Orchard-Apple}}
}

2. The Methodology Paper:

@article{jaros_methodology_2025,
  author    = {Jaros, Miroslav and Slany, Vlastimil and Skrob, Ondrej and Trenz, Oldrich and Stastny, Jiri and Samarek, Robert and Martinek, Radek},
  title     = {Methodology for Extracting Representative Frames from Video Sequences for Dataset Creation in Agricultural Applications},
  year      = {2025},
}

3. The USDA Data Source (if using data from usa_ folders):

@misc{parton_multispectral_2018,
  author       = {Parton, J. and Steward, B. and Schumacher, L. and Dowell, F.},
  title        = {Multispectral Videography Acquired from UAV during Corn and Soybean Growth},
  year         = {2018},
  publisher    = {Ag Data Commons},
  doi          = {10.15482/USDA.ADC/1416008},
  url          = {https://doi.org/10.15482/USDA.ADC/1416008}
}

Contact

Miroslav Jaros
Mendel University in Brno
Email: [email protected]

Acknowledgments

• We gratefully acknowledge the USDA Ag Data Commons for providing the public data used in parts of this dataset (DOI: 10.15482/USDA.ADC/1416008).
• This work received support from the Internal Grant System of the Faculty of Business and Economics, Mendel University in Brno, grant number IGA24-PEF-DP-005.