Cattle Body Parts Image Dataset for Object Detection
Intro
This dataset is a curated collection of images featuring various cattle body parts aimed at facilitating object detection tasks. The dataset contains a total of 428 high-quality photos, meticulously annotated with three distinct classes: "Back," "Head," and "Leg."
The dataset can be downloaded using this link. The dataset is also available at Roboflow Universe.
Acquisition
The dataset creation involved the following steps:
- Initial Data: Images were collected and annotated to create a base dataset for training.
- Model Training: A YOLOv7 model was trained to recognize target objects in the annotated images.
- Data Acquisition Script: An automated script fetched videos from the internet.
- Conversion and Filtering: Videos were turned into frames; similar frames were filtered out using Cosine Similarity.
- Object Detection: The trained model identified objects in the new images.
- Quality Check: A comprehensive review ensured dataset accuracy and consistency.
Motivation
Accurate and reliable identification of different cattle body parts is crucial for various agricultural and veterinary applications. This dataset aims to provide a valuable resource for researchers, developers, and enthusiasts working on object detection tasks involving cattle, ultimately contributing to advancements in livestock management, health monitoring, and related fields.
Data
Overview
- Total Images: 428
- Classes: Back, Head, Leg
- Annotations: Bounding boxes for each class
Below is an example image from the dataset.
Contents
📦 Cattle_Body_Parts_OD.zip
┣ 📂 images
┃ ┣ 📜 image1.jpg
┃ ┣ 📜 image2.jpg
┃ ┗ ...
┗ 📂 annotations
┣ 📜 image1.json
┣ 📜 image2.json
┗ ...
Annotation Format
Each annotation file corresponds to an image in the dataset and is formatted as per the LabelMe JSON standard. These annotations define the bounding box coordinates for each labeled body part, enabling straightforward integration into object detection pipelines.
License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Contact
For any questions, concerns, or collaboration opportunities, please don't hesitate to contact me on my LinkedIn account.