SpiceVision: An Image Classification Dataset with a Structured Knowledge Base of Bangladeshi Spices

SpiceVision is a curated image dataset designed for image classification, retrieval, and food-related computer vision research with a specific focus on Bangladeshi spices. The dataset consists of 8000+ high-resolution RGB images representing 60 commonly used spices and spice blends in Bangladesh, with each class containing 100+ images to ensure class balance.

Images were collected from local markets and household kitchens in Dhaka, Bangladesh, using multiple smartphone cameras under diverse real-world conditions. Variations include different lighting setups (daylight and indoor lighting), backgrounds (white, wooden surfaces, kitchen countertops), viewpoints (top-view, side-view, and close-up), and spice forms (whole, powdered, crushed, and blended). This diversity enhances model robustness and generalization.
Each spice class is labeled with a unique class key. containing Bangla and English names, spice category, and common culinary usage. The dataset is organized in a folder-based structure suitable for direct use in machine learning pipelines.
SpiceVision supports applications such as spice recognition, educational tools, smart kitchen systems,and retrieval-based food AI research, particularly in low-resource and multilingual contexts.