Multi-Class Waste Mapping Dataset

This dataset includes very detailed 3 cm aerial images and matching segmentation masks that label different land features. It is meant to support the automatic detection and mapping of informal waste piles. Because the images capture the ground at a very fine level of detail, they make it easier to distinguish waste from other land cover that looks similar. While the classes are labelled by their IDs, the table below provides a detailed breakdown of the land cover categories:

ID

Land Cover Class

Object Captured 

1

Waste

Waste

2

Bare Soil

Soil, Wet soil

3

Vegetation

Vegetation, Vegetation (dry)

4

Built-up Surface – Rooftop

Rooftop (silver), Rooftop (rusty), Rooftop (other color)

5

Built-up Surface – Paved Area

Tarmac, Concrete

6

Water Body

Surface water

7

Natural Surface – Rock

Rocks

8

Non-persistent Object (e.g., car)

Car, Unrecognizable white object

9

Ancillary Class (e.g., Shadow)

Shadow