ProdSynth-20

ProdSynth-20 is a synthetic image dataset designed for fine-grained product classification. It contains 50,000 cropped images, evenly distributed across 20 product categories (2,500 images per class). The dataset was generated in a virtual retail environment to highlight the intrinsic visual features of each product while maintaining intra-class variability and inter-class similarity — essential characteristics for training robust classification models. Variations in the dataset include: Changes in product orientation Subtle differences in packaging design Visual changes under different lighting conditions These aspects make ProdSynth-20 particularly suitable for training and benchmarking computer vision models in retail and inventory management scenarios.