Using image annotation to develop smart waste sorting solutions
Type of service
PICVISA is a Barcelona-based supplier of specialized optical sorting equipment for the classification of materials, recovery and valuation of waste. Their products can select, classify and pick out materials by composition, shape and color using solutions based on robotics, artificial intelligence, and machine vision.
Creating equipment that can minimize material losses with high purity and recovery rates is their primary goal, in addition to continuously offering the best solution to their customers in a constantly changing environment. Their work is a vital part of a more sustainable society where materials in today’s waste can be reused in future products instead of throwing them away.
Dani Carrero, the technical Director at PICVISA, told us:
All projects nowadays come with new classification and identification challenges, and this requires very crucial information management. It is very important to collaborate with companies like Humans in the Loop to face these challenges.
Dani Carrero, Technical Director
PICVISA has various solutions, including their top product Ecopack, an optical sorting machine that allows the automatic classification and separation of various types of materials, such as PET, PEAD, BRICK, paper and cardboard, Textile, and Film. However, in order for their machine’s AI model to accurately identify these materials, PICVISA required labeled datasets with accurately classified examples.
At first, the team annotated the data themselves but as the projects grew and became more complex, they became very time consuming and complicated to manage. They needed a tool or a company that could manage the entire annotation process, where they just had to validate the labeling afterwards. In their search for a company that could simplify the annotation process for them, PICVISA came across Humans in the Loop, which has been their preferred provider since 2018.
In the face of Humans in the Loop, PICVISA found a reliable partner for data annotation who could handle the specificities of their projects. Many of them are quite challenging because they require getting used to reviewing large piles of waste on very similarly-looking images for specific objects of interest. Some projects require the annotation team to build a deep expertise in distinguishing particular materials, packages, objects, or brands.
In order to address this, Humans in the Loop trained a dedicated team which preserves the knowledge acquired from previous batches of work and which performs better with every new batch. These annotators have developed a good understanding of the subtleties of the annotation process, such as labeling truncated or cropped objects, or handling materials which are not clear. Whether they have had to perform medical waste categorization, beverage can sorting, or End-of-live vehicle parts recycling, the dedicated Picvisa annotation team has always been up for the challenge.
A couple of more factors have contributed to the continued success in Picvisa’s collaboration with Humans in the the Loop. On the one hand, Picvisa understands that clear annotation instructions are key to obtaining high-quality labels, and they always send over detailed guidelines for each new project. On the other hand, the regular communication with the Dedicated project manager Tess in order to discuss corrections or improvements needed has also been very important.
Finally, the tool used, Hasty.ai, has enabled both organizations to collaborate more efficiently. Through the “Manual review” panel, the project supervisors are able to review instances of each class together, which considerably speeds up the quality control process and helps to detect outliers. PICVISA also has access to the Manual review option, which provides transparency and oversight on the project’s progress.
In over thirty projects, Humans in the Loop has provided over 320k polygonal annotations of various materials to help train Picvisa’s machine learning models. The datasets labelled by Humans in the Loop have contributed to the successful development of the optical sorting solutions ECOGLASS and ECOPACK.
The ECOGLASS optical sorter allows for the automatic classification and separation of objects based on their materials especially designed for glass recovery and cleaning processes in different flows, while ECOPACK can automatically classify and separate materials, with the aid of its flexible multispectral vision, and is designed to work on conveyer belts. The data labeled by Humans in the Loop has also been used for the robotics solution ECOPICK, which can replace a manual picker and can not only classify materials but also perform quality control and automatically extract those materials that fail it.
PICVISA’s Dani Carrero shares: “Humans in the Loop is a very responsible and professional company, with excellent service and a labelling team that is continuously improving.” We are looking forward to continuing our successful collaboration with PICVISA and we wish them the best of success with the global expansion of their AI-powered waste sorting solutions.