An image dataset for shopping trolley detection in computer vision with annotated YOLO-ready labels
Description
This dataset is designed for training and evaluating object detection models, specifically optimized for the detection of shopping trolleys (shopping carts) in retail, supermarket, and parking lot environments. The collection consists of 2,793 standardized images and their corresponding annotations in YOLO format. To minimize false positives and improve model robustness against complex environments, the dataset includes 250 contextual background images (empty backgrounds) with zero annotations. Key dataset statistics: • Total images: 2,793 • Images containing shopping trolleys: 2,543 • Background/Empty images: 250 • Total annotated object instances (bounding boxes): 3,571 • Image specification: 512x512 pixels, center-cropped, format PNG. • Number of classes: 1 (Class 0: 'shopping-trolley')
Files
Steps to reproduce
1. Data gathering: Initial images containing shopping trolleys were collected and annotated. To counteract model hallucinations and false positives in empty zones, 250 contextual background images of supermarket aisles, cash registers, and parking lots without target objects were integrated. 2. Automated filtering and curation: A Python pipeline was developed using the Pillow library to filter out low-resolution images (<512px). Valid images were center-cropped to a strict 512x512 resolution to maintain geometric consistency for anchor boxes. 3. Perceptual deduplication: To ensure data purity, a Perceptual Hashing (pHash) algorithm (via the imagehash library) was executed. Visual twins and duplicated files were automatically removed based on their hash signature. 4. Format standardization: All images were natively converted and saved as RGB PNG files. 5. Annotation mapping: Annotation text files (.txt) were strictly synchronized with image filenames using a sequential naming convention (from 1 to 2793). Empty .txt files were mirrored for the 250 background images to comply with YOLO detection standards.
Institutions
- Universidad de SalamancaCastille and León, Salamanca