2021
Waste detection
Detect & classify litter in the wild — merged benchmarks, open models, a living dataset index, and community hackathons in Pomerania.
Prior waste datasets were often small, class‑imbalanced, or tied to one domain (e.g. aerial only). The detect‑waste benchmark unifies detection boxes into one “litter” taxonomy across tens of thousands of images—indoor, outdoor urban/natural, and underwater scenes, varied lighting and camera gear—so models generalize beyond a single collection.
Classify‑waste
A parallel classification benchmark aggregates public datasets into eight labels aligned with Gdańsk recycling rules—bio, glass, metals & plastics, paper, non‑recyclable, other (construction, e‑waste, hazardous), highly degraded “unknown” waste, and a true background class—reflecting messy real‑world streams rather than tidy lab subsets.
Community programme
A months‑long WiMLDS educational track brought Tri‑City participants together with mentors from industry and research to train modern detectors on shared baselines, publish code, and release the extended benchmarks that later supported the Waste Management journal article.
Waste datasets review
A separate, community‑indexed GitHub list catalogs image datasets for litter, garbage, and waste detection and classification—useful when you are comparing domains, licenses, and label granularity before training or benchmarking.
Hack4Environment
Co‑organized with WiMLDS Trójmiasto and the Digital Innovation Hub DIH4.AI: a hackathon on waste, environment monitoring, and AI literacy—hands‑on activities across age groups, from environmental storytelling to monitoring ideas, in the same AI4Good thread as Detect Waste.
Plastic waste and population pressure make civic awareness as important as model accuracy; the event aimed to shape attitudes toward land care alongside technical prototypes.