Pharmassist Dataset
Description
The dataset consists of 4,997 images belonging to 53 tablet classes. The dataset contains both original photographs and augmented images. A total of 2,050 images are original (raw) photographs, while 2,947 images are augmented samples derived from the original dataset. The 53 classes correspond to different tablet types represented by their respective blister-pack appearance. The dataset is intended as a multi-class image classification resource, where the objective is to associate a torn or partially used blister pack with its corresponding tablet class. Used tablet blister packs were obtained from personal and voluntary sources and subsequently photographed. Data collection was conducted over a period of four months, from September 2025 to December 2025, at multiple locations across Chennai, India. The images were intentionally photographed under varying visual conditions to increase the diversity of the dataset. Tablet blister packs were photographed from multiple viewing angles and orientations, rather than using a fixed camera position. Variations in illumination and background were also introduced naturally during the collection process. These variations are intended to approximate the conditions under which a tablet blister may be shot in a practical medication-identification scenario. The collection process therefore incorporates variations arising from: • Different viewing and camera angles; • Different orientations of the blister pack; • Variations in ambient lighting conditions; • Different image backgrounds;
Files
Institutions
- Vellore Institute of Technology UniversityTamil Nadu, Vellore