Dal Lake Floating Plastic Waste Detection Dataset (FloPWD 2025)

To support semantic and instance segmentation tasks, the Dal Lake Floating Plastic Waste Detection Dataset (FloPWD 2025) includes four specialized files: Color_ID, JSON, TrainLabel_ID, and Instance_ID. The Color_ID file defines RGB color codes for each class (plastic waste, background) used in visualizing segmentation outputs. The JSON file contains polygonal annotations created using the VGG Image Annotator to specify the exact pixel regions corresponding to plastic waste. These annotations are further converted into TrainLabel_ID files, which assign unique integer class labels to each pixel to form the ground truth for semantic segmentation models. The Instance_ID file distinguishes between individual plastic waste instances within an image to support instance-level segmentation. Together, these files form the core data required for training and evaluating deep learning models in plastic waste detection at a pixel and object level.

Supplementary Files Description(s)

(i) Color_ID: This field maps each class (plastic waste [class 1] and background [class 2]) to a specific RGB color code. It is used primarily for visualization and post-processing of segmentation results.

(ii) JSON: This file contains the polygonal annotation coordinates (pixel-level) for plastic waste regions in each image, created using the VGG Image Annotator. These annotations are essential for generating ground truth labels such as TrainLabelID.

(iii) TrainLabel_ID: Represents the labeled pixel values used as ground truth for training semantic segmentation models. Each pixel is assigned an integer class ID based on the JSON annotations.

(iv) Instance_ID: Used in instance segmentation tasks to distinguish between different plastic waste objects within the same image, allowing models to segment multiple instances individually.