WADE: A Waste Annotation and Reasoning Chain Dataset for Waterbody Monitoring in Bangladesh
Description
WADE (Waste Annotation and Reasoning Chain Dataset for Waterbody Monitoring in Bangladesh) is an open-access multimodal dataset curated for artificial intelligence, computer vision, vision-language models (VLMs), and environmental monitoring research. It targets real-world surface water pollution, invasive aquatic vegetation, and eutrophication across freshwater bodies (stagnant ponds, slow-moving canals, rivers, and lakes) in Bangladesh. Unlike conventional object detection datasets that only provide spatial bounding boxes, WADE introduces Fine-grained Chain of Thought (FCoT) visual reasoning for each grounded instance. Each annotation pairs spatial coordinates with rigorous diagnostic rationale—including perceptual observations, contrastive disambiguation rules, fallback guidelines, and environmental conclusions by enabling both standard object detectors and reasoning-driven Multimodal Large Language Models (MLLMs).
Files
Institutions
- United International UniversityDhaka Division, Dhaka