InstruPhys: A Real-World Benchmark Dataset for Deep Learning-Based Detection, Classification, and Recognition of Physics Laboratory Instruments in Educational Environments.

Description

The PhysiSens dataset comprises 3,048 raw images of physics instruments collected from the Physics Lab Support Room at United International University (UIU). Following annotation, the dataset was organized into 20 distinct classes representing commonly used physics laboratory instruments. The annotated images were divided into training (70%, 2,134 images), validation (20%, 610 images), and testing (10%, 304 images) subsets to support model development, optimization, and performance evaluation. To ensure consistency, all images underwent automatic orientation correction and were resized to 640 × 640 pixels using the stretch method. No additional data augmentation techniques were applied in this dataset version. Images were captured under varying lighting conditions, against diverse backgrounds, and from multiple viewing angles to represent realistic laboratory environments and improve the applicability of trained models to real-world physics instrument detection tasks.

Steps to reproduce

For your Roboflow Physics Lab Apparatus dataset, you can use the following steps: 1. Download the dataset from Roboflow in YOLO format. 2. Extract the dataset into /content/physics_lab_apparatus-11. 3. Verify that the dataset contains train, valid, and test directories, along with the data.yaml configuration file. 4. Install the required Python dependencies (PyYAML and Pillow). 5. Load the data.yaml file to retrieve the dataset class names. 6. Scan the training, validation, and testing image directories. 7. Match each image with its corresponding YOLO annotation (.txt) file. 8. Extract image dimensions, image format, class IDs, class names, and bounding-box counts. 9. Combine the extracted information into a structured metadata.csv file. 10. Validate the generated metadata and verify the image counts for each dataset split.

Institutions

• United International University

  Dhaka District, Dhaka

Categories

Licence