wang4067/green_stormwater_infrastructure_maintenance_dataset

Green Stormwater Infrastructure Maintenance Dataset

Repository: wang4067/green_stormwater_infrastructure_maintenance_dataset

Overview

The Green Stormwater Infrastructure (GSI) Maintenance Dataset is a visual dataset designed to support research on image-based diagnosis and maintenance decision-making for urban stormwater systems. The dataset focuses on real-world conditions commonly encountered by field workers, rather than idealized or design-stage representations.

All images are collected from publicly accessible web sources and represent practical maintenance scenarios in urban environments.

---

Dataset Structure

The dataset is organized into four folders, each representing a common stormwater-related field scenario:

1. Water accumulation on the road (36 images)

Images showing surface water ponding on roads or paved areas, often indicating:

• blocked or undersized drainage,
• clogged stormwater inlets,
• reduced infiltration capacity of permeable surfaces.

These scenes reflect early warning signs of stormwater system malfunction after rainfall.

---

2. Rain garden (85 images)

Images of rain gardens and similar bioretention features under various conditions, including:

• normal operation,
• standing water,
• overgrown or dead vegetation,
• signs of soil compaction or poor drainage.

This folder supports visual assessment of small-scale green stormwater infrastructure performance and maintenance needs.

---

3. Stormwater after heavy rain (39 images)

Images captured shortly after heavy rainfall events, showing:

• temporary flooding,
• overflow conditions,
• surface runoff accumulation in urban areas.

These images provide context for stress conditions under which maintenance issues become visible.

---

4. Stormwater maintenance work (85 images)

Images depicting maintenance activities performed by field workers, such as:

• debris removal,
• inlet cleaning,
• surface cleaning,
• routine inspection tasks.

This folder offers real-world visual references for actionable maintenance operations.

---

Intended Use

This dataset is intended for:

• multimodal and vision-language models,
• image-based stormwater failure identification,
• decision-oriented maintenance planning,
• research on AI-assisted urban infrastructure management.

The dataset is not designed for object detection or fine-grained segmentation, but rather for high-level visual reasoning and maintenance decision support.

---

Data Collection and Ethics

Images are collected via keyword-based search from publicly accessible web sources and are used solely for non-commercial academic research. Original images are not redistributed beyond this dataset repository when restricted, and usage is intended to be transformative in nature.

---

Limitations

• Images vary in quality, angle, and lighting.
• Labels are coarse-grained and scenario-based.
• Ground-truth engineering measurements are not provided.

Despite these limitations, the dataset captures realistic conditions encountered in day-to-day stormwater infrastructure maintenance.

---

license: apache-2.0