PlasticVision-BD: A Benchmark Dataset for Plastic Waste Detection

Description

PlasticWaste-BD is a real-world annotated image dataset developed for automatic plastic waste detection and classification using computer vision and deep learning techniques. The dataset comprises high-quality images collected under diverse real-world environmental conditions, including variations in illumination, viewing angles, backgrounds, and object orientations, to reflect practical waste management scenarios. The dataset contains five common categories of plastic waste: Plastic-Bottle, Plastic-Cup, Plastic-Packet, Plastic-Straw, and Polythene. Every object instance has been manually annotated with bounding boxes following a rigorous annotation protocol to ensure accurate localization and consistent class labeling. Multiple rounds of quality verification were performed to minimize annotation errors and improve dataset reliability. PlasticWaste-BD is designed to support research in object detection, image classification, waste recognition, environmental monitoring, and intelligent recycling systems. The dataset is compatible with widely used deep learning frameworks, including YOLO, RT-DETR, Faster R-CNN, SSD, EfficientDet, and other modern computer vision models. It can also serve as a benchmark for evaluating novel detection algorithms and transfer learning approaches. The repository includes the complete annotated dataset, annotation files, dataset configuration files, metadata, and documentation necessary for reproducible research.

Steps to reproduce

Steps to Reproduce : 1. Download the PlasticWaste-BD dataset from the repository and extract the compressed archive. 2. Organize the dataset according to the provided directory structure (e.g., train, valid, and test folders). 3. Install the required deep learning framework (e.g., Ultralytics YOLO, RT-DETR, or another supported object detection framework) and its dependencies. 4. Use the provided dataset configuration file (data.yaml) to load the dataset. 5. Train an object detection model using the recommended training settings or adjust the hyperparameters according to the experimental requirements. 6. Evaluate the trained model on the independent test set using standard object detection metrics, including Precision, Recall, mAP@0.50, and mAP@0.50:0.95. 7. The dataset can also be converted into COCO, Pascal VOC, or other supported annotation formats for use with alternative computer vision frameworks. Recommended Environment : Python 3.10 or later; PyTorch 2.x; Ultralytics YOLO (latest stable version); CUDA-enabled GPU (recommended for training)

Institutions

• United International University

  Dhaka Division, Dhaka
• Southeast University

  Dhaka Division, Dhaka

Categories

Licence