Datasets:
license: cc-by-4.0
language:
- en
tags:
- computer-vision
- retail
- retail-imagery
- stores
- object-detection
size_categories:
- n<1K
task_categories:
- object-detection
Dataset Description
The Retail Object Detector dataset supports training and evaluating computer vision models for general object detection in retail environments. It contains 749 labeled .PNG images across train, dev, and test splits, combining real retail images with AI generated synthetic images to improve model robustness.
Each image includes a corresponding Pascal VOC XML annotation with bounding boxes and a single class label 'obj', intended for research, model development, and inventory monitoring use cases, performace may vary with lighting, occlusion, camera angle, and image quality.
Intended Use
This dataset is intended for research, experimentation, educational applications, and early-stage development of retail computer vision systems. It can be used for developing and evaluating object detection models focused on identifying and localizing items in shelf-like environments, with potential applications including shelf item detection, stock visibility analysis, and planogram compliance experiments.
Dataset Strengths
The dataset combines synthetic and non-synthetic imagery, providing two different sources of visual data for experimentation. Images represent multiple shelf and storage configurations, while maintaining a consistent object-detection annotation structure. The dataset uses a single OBJ class, allowing it to be used as a general-purpose object localization dataset without requiring product-specific classification labels.
Dataset formatting
Image formatting
Citation
@dataset{ratio1_drone_dataset_2026,
author = {Ratio1},
title = {Retail Object Detection},
year = {2026},
publisher = {Hugging Face},
url = {https://huggingface.co/datasets/ratio1/Retail_Object_Detection},
version = {v1.0}
}