Dataset Taxonomy


Trash classification

Waste sorting is more complex than just assigning the material labels: Plastic, Glass, Metal and Paper or as Gervasio shows.


Certain objects are in fact composites (e.g. Tetrapak) made of plastic, paper and aluminium, plus there are 7 broad classes of plastics, including 'bioplastics'. While these may seem hard to distinguish visually, we can actually guess the type of plastic based on the object type. E.g. Water bottles are generally made of PET, whereas opaque detergent bottles are made of HDPE and Tupperwares are made of PP. Therefore, trash is classified by both object and material using the categories and super categories below. Description is provided for less obvious categories. N.b. Categories inherit the supercategory's name if these do not branch.

Super category Category Notes
Aluminium foil Aluminium foil -
Battery Battery -
Blister pack Aluminium blister pack Containers used to store capsules (e.g. pills)
Blister pack Carded blister pack Paper-back package
Bottle Clear plastic bottle Water and soft drink bottles made of PET
Bottle Glass bottle Includes beer and wine bottles
Bottle Other plastic bottle Opaque or translucent. Generally made of HDPE. Includes detergent bottles
Bottle cap Plastic bottle cap -
Bottle cap Metal bottle cap -
Broken glass Broken glass -
Can Aerosol -
Can Drink can Aluminium soda can
Can Food can Steel can
Carton Corrugated carton Includes cardboard boxes
Carton Drink carton Tetrapak composites
Carton Egg carton -
Carton Meal carton Includes sandwich boxes, paper plates, take-out boxes
Carton Pizza box -
Carton Toilet tube -
Carton Other carton Paperboard boxes
Cigarette Cigarette Cigarette butts
Cup Paper cup -
Cup Disposable plastic cup Generally made of PET
Cup Foam cup Polystyrene Cup
Cup Glass cup -
Cup Other plastic cup Reusable plastic cups, thicker than disposable ones
Food waste Food waste -
Glass jar Glass jar -
Lid Plastic lid Includes cup lids
Lid Metal lid Generally glass jar lids
Paper Normal paper -
Paper Tissues -
Paper Wrapping paper -
Paper Magazine paper Plastified paper used in catalogues
Paper bag Paper bag Brown bag
Paper bag Plastified paper bag Bakery bags that come with transparent film
Plastic bag & wrapper Garbage bag -
Plastic bag & wrapper Single-use carrier bag -
Plastic bag & wrapper Polypropylene bag Reusable bags
Plastic bag & wrapper Plastic Film May be transparent or opaque. Inludes bread bags, cereal bags and produce bags
Plastic bag & wrapper Six pack rings -
Plastic bag & wrapper Crisp packet So common that it needs its own category
Plastic bag & wrapper Other plastic wrapper Can be made of aluminium. Includes candy wrappers, retort pouches and yoghurt lids
Plastic container Spread tub Includes margarine tubs and yoghurt pots
Plastic container Tupperware HDPE microwavable tub
Plastic container Disposable food container Includes black trays and PET containers
Plastic container Foam food container Styrofoam takeaway boxes
Plastic container Other plastic container -
Plastic glooves Plastic glooves -
Plastic utensils Plastic utensils -
Pop tab Pop tab -
Rope Rope Includes fishing nets
Scrap metal Scrap metal Includes all metal except cans
Shoe Shoe -
Squeezable tube Squeezable tube Includes toothpaste and glue tubes
Straw Plastic straw -
Straw Paper straw -
Styrofoam piece Styrofoam piece -
Other plastic Other plastic Includes other objects or fragments made of plastic
Unlabeled litter Unlabeled litter Unknown object of unknown material. Any ambiguous object.

Background classification

Scenes are labeled using the following tags:

  • Pavement, roads
  • Vegetation
  • Sand, dirt, stones
  • Water
  • Trash
  • Indoor, man-made
  • Clean

These are not mutually exclusive since one photo may contain different backgrounds.

Context classification

Objects are also distinguished between litter and non-litter based on the scene context

Class Description
Litter Mostly trash on the floor
Not litter Objects in our class system that are (i) not trash yet or (ii) in a garbage bag or (iii) handheld
Unknown In certain images, object context is ambiguous