MTA (Malaysian Trash Annotation): In-The-Wild Instance Segmentation Dataset for Waste Segregation

Malaysian Trash Annotation or MTA in short is a localised instance segmentation dataset designed to address the current issue of waste segregation in Southeast Asia context. Compare with other openly-sourced dataset, MTA is heavily designed around real-world situation where it provides polygon masking to capture the exact geometry of the trash. Goal of the project:To provide COCO Segmentation formatted database that capable of distinguish between chemically distinct but visually similar material (PET material and Thermoform) and estimating position of garbage in uncontrolled situation. Standard Operation Procedure of Data Annotation: Dataset Statistics (As of Version: 11122025) Task: Instance Segmentation (Polygons) Classes: 28 (Material-based separation) Environment: Supermarkets (clean and ideal condition), Roadside (weathered), Food Courts (occluded/deformed). Hardware: Heterogeneous capture using 5 different devices (Oppo, Honor, Samsung, iPhone, Realme). Capture Protocol: Aspect Ratio locked at 4:3, Multi-angle (Top, Side, 45-degree) with no AI enhancements and no HDR enabled. Classes of Dataset (As of Version: 11122025)100 Series: Plastics 101_PET_Bottle: Clear/Tinted transparent bottles (Mineral water, Coke, 5kg Oil bottles). 102_HDPE_Plastic: Opaque rigid jugs (Shampoo, Detergent, Milk bottles). 103_Disposable_Cup: Thin plastic cups (Boba/Coffee). Labeled only if empty/dry. 104_Soft_Plastic: Stretchy LDPE films, courier bags, bubble wrap. 105_Plastic_Container: Clear rigid thermoforms (Egg trays, Fruit boxes). 106_Plastic_Cutlery: Disposable spoons, forks, and knives. 200 Series: Metals 201_Aluminium_Tin: Carbonated drink cans (Smooth wall, non-magnetic). 202_Tin_Can: Food cans (Sardine/Soup, ribbed wall, magnetic). 203_Spray_Can: Aerosol cans (Paint, Insecticide). 300 Series: Paper & Cardboard 301_Cardboard: Corrugated boxes (Delivery parcels) and heavy paperboard. 302_Paper: Flat paper sheets, newspapers, A4. 303_Tetra_Pek: Laminated drink cartons (Milo, Yeos, Juice boxes) with silver lining. 400 Series: Glass & Ceramic 401_Glass: Jars, Sauce bottles, Drink bottles. 402_Ceramic: Broken plates, mugs, flower pots 500 Series: Contamination & General Waste 501_Contaminated: Packaging (Boxes/Bowls) that contains oil, sauce, or liquid residue. 502_General_Trash: Tissues, dirt, small debris, and non-recyclable composites. 503_Straw: Drinking straws (Plastic and Paper). 504_Receipt: Thermal paper (Shiny/Coated). 505_Food_Wrapper: Laminated foil wrappers (Chips, Candy, Maggi). 506_Wooden_Stick: Satay sticks, chopsticks, skewers. 512_Cigarette_Butt: Used filters. 513_Mask: Disposable face masks (Biohazard). 514_Ikat_Tepi: Plastic drink bag tied with string, containing liquid. 515_Food_Waste: Organic leftovers (Rice, Bones, Fruit skins). Distinct from the container. 600 Series: E-Waste (Hazardous) 601_Electronic_Device: Vapes, Powerbanks, Old Phones. 602_Battery: AA/AAA, Button cells (Pending collection). 603_Cable_Wire: USB cables, Chargers (Pending collection). 604_Textile: Fabric scraps, clothes. Devices (As of Version: 08122025) Oppo Reno 14 MY Version (OppoReno14) Samsung Note 10 MY Version (SamsungNote10) Iphone 11 MY Version (Iphone11) Honor 200 MY Version (Honor200) Realme 5s MY Version (Realme5s) Test Result: Notes: Reflections: Contains samples on glass/steel tables; polygons are tightly cropped to the object, excluding ghost reflections. Deformation: Includes crushed cans and flattened bottles to simulate end-of-life object states. Localization: Features "Ikat Tepi" and "Keropok" wrappers specific to the Malaysian region.