After five months and around 1000 (wo)man-hours of uphill battle we successfully finished our project. The “detect waste” will be a memorable experience for many of us. We finished with over 80 commits to the official repository, 500 experiments in 5 projects on neptune.ai (and untracked numerous trainings), thousands of lines of code.

Summary of the project

We proposed a two-stage ML model and provided extensive review of datasets and approaches in litter detection. Check out the summary of our project Creating an AI to Combat Environmental Pollution written by Magdalena Kortas and Sylwia Majchrowska on towards data science. We additionally publish our results on arxiv: Waste detection in Pomerania: non-profit project for detecting waste in environment, authored by: Sylwia Majchrowska, Agnieszka Mikołajczyk, Maria Ferlin, Zuzanna Klawikowska, Marta A. Plantykow, Arkadiusz Kwasigroch, Karol Majek. Our github repository: https://github.com/wimlds-trojmiasto/detect-waste

Our blog

Here is the list of all of our twelve blog posts:

• Introduction - How a detectwaste.ml story begins
• Exploratory Data analisys - Learn about TACO bboxes
• Hack4Environment - Event summary - How 17 teams worked together for 24h
• Data augmentation - the beginning - The introduction to data augmentation
• Data augmentation - next steps - Extension of data augmentation techniques
• How scientists use artificial intelligence to identify garbage? - Brief review of state-of-the-art papers
• Coming back to the basics - AI Theory - To understand some of the deeper AI concepts, you need to understand its basics
• Basic metrics to compare waste detection models - Evaluation metrics that data scientists should know
• Comparing waste detection approaches on extended TACO dataset - Differentiating waste instances under a single class label is challenging task
• One to rule them all - combination of few datasets of waste - Neural networks in the garbage detection service
• Detect waste with Transformer - Transformers in the garbage detection service
• Pseudo-labeling as a boost in waste classification task - Waste sorting is the process by which waste is separated into different categories

Project roles

Team members:

Nine out of ten women stayed with us to the end! Each one of them got appointed the role title adequate to the real project role. Congratulations!

• Agnieszka Mikołajczyk - Machine learning Team Leader
• Sylwia Majchrowska - Lead Machine learning Researcher
• Maria Ferlin - Machine learning Researcher
• Ewa Marczewska - Machine learning Researcher
• Zuzanna Klawikowska - Machine learning Engineer
• Marta Plantykow - Machine learning Engineer
• Magdalena Kortas - Machine learning Engineer
• Anna Brodecka - Data Scientist
• Katarzyna Łagocka - Junior Data Scientist

Organizers of the project:

• Magdalena Kortas - Project manager
• Ewa Marczewska - Project manager
• Agnieszka Mikołajczyk - Project manager

How to cite us

You can read about our results on arxiv: Waste detection in Pomerania: non-profit project for detecting waste in environment, authored by: Sylwia Majchrowska, Agnieszka Mikołajczyk, Maria Ferlin, Zuzanna Klawikowska, Marta A. Plantykow, Arkadiusz Kwasigroch, Karol Majek.

@misc{majchrowska2021waste, title={Waste detection in Pomerania: non-profit project for detecting waste in environment}, author={Sylwia Majchrowska and Agnieszka Mikołajczyk and Maria Ferlin and Zuzanna Klawikowska and Marta A. Plantykow and Arkadiusz Kwasigroch and Karol Majek}, year={2021}, eprint={2105.06808}, archivePrefix={arXiv}, primaryClass={cs.CV} }

Code: https://github.com/wimlds-trojmiasto/detect-waste