ResiDual Dataset
This page describes the test dataset related to the paper 'Identifying Impurities in Liquids ofPharmaceutical Vials' presented at the 27th International Conference on Pattern Recognition (ICPR). The dataset here described has been acquired by Performing Beyond Limits (PBL) automation industry. The dataset consists of sequences of frames depicting vials filled with pharmaceutical liquid. Some of these vials are free of impurities, while others contain traces of rubber, glass, sand, and plastic.
Features
The table below resumes the core information about the dataset.
For a complete technical description please refer to the our paper.
The dataset contains annotations for five different classes. One class represents clean vials, indicating the absence of any sort of impurities. The other four classes refer to different types of residual particles: plastic particles, rubber or silicone particulates, glass pieces of various sizes, and sand debris.
| Field | Value |
|---|---|
| File Format | JPEG |
| Dataset Classes | 5 |
| Vial Illumination | LED |
| Number of Sequences | 730 |
| Images Acquisition Cam | Matrix Vision Camera |
| Total Number of Images | 13.87 |
| Number of Frames per Sequence | 19 |
| Original Images Size (Height, Width) | 1600, 768 |
Dataset Structure
The ResiDual dataset follows the DSMIL dataset format. Hence, the dataset is divided into 5 subfolders, each representing a different class: Clean, Sand, Rubber, Plastic, and Glass. Within each subfolder, there are multiple subfolders, each containing 19 frames of the same vial.
