From December 5th to 9th 2022, a team of 4 Data Scientists from the DataLab Groupe has participated to an international challenge, jointly organized by the Ocean Cleanup NGO and the companies: Kili, Weights & Biases and OVHcloud. The objective was to contribute to the development of Artificial Intelligence systems fighting against plastic pollution in the marine environment. The DataLab Groupe team finished the challenge on the podium among 300 other teams from companies and laboratories worldwide.

Plastic pollution in the marine environment is a global threat to the health of marine species, human health, food safety, coastal tourism, etc. Indeed, among the 350 million tons of plastic produced each year, it is estimated that more than 15 million tons end up in the ocean.

The fight against this type of pollution is one of Crédit Agricole’s commitments with, for example, the Plastic Odyssey initiative, of which the Group has been a partner since the project’s prototyping phase in 2018.

Artificial intelligence can play an important role by helping to spot plastic waste more quickly from images taken on boats, for example.

Thus, between December 5 and 9, an international challenge was jointly organized by the Ocean Cleanup NGO and the companies: Kili, Weights & Biases and OVHcloud. It brought together 300 teams from companies and research laboratories around the world. The objective was to create a reference dataset for learning and testing models for detecting plastic waste in the open ocean. The images of the dataset have been taken on The Ocean Cleanup’s boats.

The challenge consisted of two independent tasks to detect plastic waste objects in a dataset of over 180,000 images:

  • Task 1: manual/semi-manual annotation;
  • Task 2: automatic annotation model.

After evaluating each team’s contribution, organizers revealed that the DataLab Groupe team won 2nd and 3rd place respectively on these two tasks.

What is the DataLab Groupe’s recipe?

Thanks to the implementation of an active learning process, consisting in building an AI model trained on the most relevant images, iteratively selected and labeled manually to characterize the presence or absence of the sought objects, the team was the only one to make its place on the podium on both tasks.

A big congratulations to this winning team:

  • Aghiles Azzoug
  • Achraf Saghe
  • Paul Wassermann
  • Saif Fares