Dataset of Annotated Tesla and Non-Tesla Vehicle Images Acquired Under Variable Lighting Conditions

<jats:p>This article describes a labeled image dataset of Tesla and non-Tesla vehicles collected under varying real-world lighting conditions in the city of Haugesund, Norway. In some artificially illuminated environments, in-camera exposure compensation and automatic night mode were used during image acquisition to increase variation in image brightness and visibility. The dataset was developed to support computer vision research on vehicle classification in realistic and challenging illumination scenarios, with relevance for multi-attribute visual vehicle recognition. Images were captured using mobile phones in public parking lots, parking garages, and along municipal and county roads, representing environments where vehicles are commonly observed in everyday traffic and parking situations. The dataset includes variation in vehicle type, Tesla model, production period, color, camera angle, camera distance, surrounding environment, and lighting conditions. The dataset contains image-level lighting-condition labels categorized as light, medium, and dark, enabling evaluation of vehicle classification performance across different illumination scenarios. The images were originally acquired in Apple’s HEIC format and converted to JPG/JPEG format to improve compatibility with annotation tools, image processing software, and machine learning frameworks. No image enhancement was applied during format conversion; the only subsequent image editing involved masking visible license plates and individuals to protect privacy. The dataset can be used to train, evaluate, and benchmark deep learning models for Tesla versus non-Tesla classification, Tesla model classification, and production-period classification. It is also suitable for studying how lighting variation, dataset design, and training strategies influence prediction accuracy in fine-grained vehicle recognition tasks.</jats:p>