KNR – PET –7 : An industrial – grade image dataset of deformed and contaminated PET bottles for fine-grained sorting in emerging markets
Description
The KNR – PET –7 dataset comprises over 5,000 high-resolution RGB images of post-consumer Polyethylene Terephthalate (PET) plastic bottles, captured directly at an operational industrial recycling facility in Vietnam. KNR – PET –7 is specifically designed to reflect the in-the-wild complexities of waste streams in emerging markets, featuring severe morphological deformations (crushed, flattened) and heavy contamination (mud, residual liquids). The dataset is meticulously annotated for object detection tasks using the standard YOLO bounding box format (.txt files). It features 7 fine-grained industrial PET categories: B1, B2.1, B2.2, B3, B4, B5 and B6. This dataset serves as a robust and challenging benchmark for developing real-time machine vision and AI-driven robotic sorting systems in circular economy initiatives.
Files
Steps to reproduce
To utilize the KNR – PET –7 dataset for training and evaluating object detection models, follow these steps: Download and Extract: Download the provided dataset archive (KNR – PET –7_dataset.zip) and extract it. The internal folder structure is strictly organized into images/ and labels/ directories, following standard deep learning practices. Check Annotations: Annotations are provided in YOLO format (.txt). Each text file corresponds to an image and contains bounding box coordinates mapped sequentially via five normalized attributes: class_id, x_center, y_center, width, and height. Class Mapping: Refer to the included README.md file to map the integer class_id (0 to 6) to their specific fine-grained PET categories (0: B1, 1: B2.1, 2: B2.2, 3: B3, 4: B4, 5: B5, 6: B6).
Institutions
- Ho Chi Minh City University of TechnologyHo Chi Minh, Ho Chi Minh City