VegQual: 2,032 Images of Fresh and Defective Vegetables for Object Detection

A dataset of 2,032 raw images depicting seven common vegetable species, each categorized into fresh and defective quality classes, resulting in 14 object categories. The images were captured under real-world conditions with variations in angles, backgrounds, and lighting, and are annotated with 4,736 bounding boxes in YOLO format. Created by Tarek Rahman and last updated in March 2026, the data is pre-split into training, validation, and test sets.

Use Cases

• Train object detection models for real-time vegetable classification based on the annotated bounding boxes.
• Develop defect detection systems for agricultural automation based on the 'fresh and defective' quality labels.
• Benchmark model performance on diverse real-world images based on the described variations in viewing angles, backgrounds, and lighting.
• Support research in food quality assessment and sustainable production based on the described categories of vegetable species.

Strengths

• Contains 2,032 raw images annotated with 4,736 bounding box instances.
• Images were captured under diverse real-world conditions, including variations in viewing angles, backgrounds, and lighting.
• Dataset is pre-processed and split into training (1,422 images), validation (406 images), and testing (204 images) subsets.
• Annotations were created using the Roboflow platform, which suggests consistent labeling standards.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• The geographic origin of the images is not specified, which may limit understanding of potential bias.
• Row count is unknown, which may limit suitability assessment for certain tasks.

Provenance

Source

figshare

Collection Method

Images were captured under real-world conditions and annotated using the Roboflow platform.

Freshness

Last updated 2026-03-29 05:04:19; freshness should be verified.