VERIDIS Dataset

Overview

This repository contains the VERIDIS dataset, a collection of annotated agricultural images for crop detection and identification. The dataset comprises field images of beet and corn crops captured by a ground-level robotic platform, organized in YOLO format for object detection tasks.

The dataset primarily captures crops at early growth stages, which is particularly relevant for applications such as plant detection, early monitoring, and precision weeding.

The dataset includes 10080 total files distributed across training (9104 files), validation (493 files), and test (483 files) sets, with images of two crop types: beet (Class 0) and corn (Class 1).

Repository Contents

Main Dataset

• veridis.zip
  • Main dataset containing the complete collection of data for crop detection and agricultural robotics research
  • Recently updated with the latest version of the dataset
  • Format: Compressed ZIP archive

Processing Scripts

1. anonymize_persons.py

Script designed for anonymizing personal data within the dataset. It implements privacy-preserving techniques to protect personally identifiable information while maintaining data utility for research purposes.

After the automated anonymization process, a manual inspection step was performed to verify that all individuals were properly anonymized, ensuring compliance with privacy requirements.

2. data_augmentation.py

Data augmentation script that expands the dataset through systematic transformations and variations generation.

3. extract_frames.py

Tool for extracting individual frames from video sequences or temporal data.

4. train.py

Main training script for machine learning and deep learning models on the VERIDIS dataset.

Dataset Splitting Considerations

The dataset is split into training, validation, and test subsets using a frame-level random strategy. As a result, frames from the same traversal or crop row may appear across different subsets.

Since frames were extracted at approximately 1 frame per second, consecutive images may exhibit high visual similarity. This can introduce temporal correlation between subsets and may lead to optimistic performance estimates when training and evaluating machine learning models.

Camera Calibration

Intrinsic and extrinsic parameters for the Intel RealSense D455 camera are provided in the file realsense_d455_calibration.txt. These parameters allow researchers to project YOLO 2D annotations into 3D space using the camera's depth data.

License

This project is licensed under CC-BY-SA-4.0 (Creative Commons Attribution-ShareAlike 4.0 International).