Multi-view RGB image dataset of four potato varieties with visually assessed scab and dry-rot severity

Description

This dataset contains 9,106 RGB JPG images of potato tubers representing four varieties: Rumana, BARI Alu-25 (Asterix), Lal Pakri, and BARI Alu-7 (Diamant). The images are organized into 18 combined variety-condition/severity classes. Images were collected at New Kafela Cold Storage, Shibganj, Bogura, Bangladesh, under natural daylight using iQOO Neo 11 and realme Narzo 50 smartphones at an approximate camera-to-tuber distance of 15-20 cm. Each physical tuber was photographed from five orientations (top, bottom, middle, left, and right) on two backgrounds (white paper and concrete floor). Source photographs were captured at approximately 4096 × 3072 pixels in JPG format. For repository release, the retained images were corrected for EXIF orientation, converted to RGB, resized while preserving aspect ratio using Lanczos resampling, padded, and standardized to 512 × 512 pixels. Disease-condition and severity labels were assigned manually through visual phenotypic assessment. Rumana images are organized into Healthy, Scab-Mild, Scab-Moderate, and Scab-Severe categories. BARI Alu-25 (Asterix), Lal Pakri, and BARI Alu-7 (Diamant) contain Healthy, DryRot-Mild, DryRot-Moderate, and DryRot-Severe categories, with Fully-Rotten categories additionally included for Asterix and Diamant. The disease and severity annotations represent visually assessed phenotypic categories and were not confirmed through laboratory pathogen testing.

Steps to reproduce

Source potato-tuber photographs were captured in JPG format at approximately 4096 × 3072 pixels using iQOO Neo 11 and realme Narzo 50 smartphones under natural daylight. Images were acquired from five orientations (top, bottom, middle, left, and right) using two backgrounds: white paper and concrete floor. For the public repository release, EXIF orientation was corrected and each image was converted to RGB. Images were resized using Lanczos resampling while preserving the original aspect ratio and were centered on a 512 × 512 canvas. Padding was added where necessary to obtain an exact output size of 512 × 512 pixels without geometrically stretching the tuber. Consequently, some images contain uniform padding bands along the horizontal or vertical edges. The processed images were saved as high-quality JPG files and retained in their corresponding combined variety-condition/severity folders. Class-wise image counts were verified after processing. The final dataset contains 9,106 images organized into 18 class folders representing four potato varieties.

Institutions

• American International University-Bangladesh

  Dhaka Division, Dhaka
• Hajee Mohammad Danesh Science and Technology University

  Rangpur Division, Dinajpur
• Jahangirnagar University

  Dhaka Division, Dhaka

Categories

Licence