Tuberculosis Autofocus Datasets

Description

This dataset provides a curated and class-balanced collection of digital microscopy images for focus-quality classification. The dataset contains 6,405 images labeled into three focus categories, Blur, Medium, and Focused, and is pre-split into training (4,401 images), validation (914 images), and test (1,090 images) sets with near-equal class distribution across all splits. Images were sampled from a larger raw collection of angular focus-scanning sessions spanning multiple fields of view, preserving natural variation in illumination, sharpness transition, and sensor response. No synthetic image enhancement, color normalization, contrast adjustment, or artificial augmentation was applied to the released images. This makes the dataset suitable for training and evaluating focus-quality classification and autofocus estimation models under realistic microscopy imaging conditions.

Institutions

• Politeknik Elektronika Negeri Surabaya

  East Java, Surabaya
• Sepuluh Nopember Institute of Technology

  East Java, Surabaya

Categories

Licence