Data labelling for image recognition

DataMondial is working with a rapidly growing European AI company. The client develops and implements self-learning systems in various sectors that help companies operate more efficiently. For one of their projects in the agricultural sector, in which developing an image recognition system is key, we have helped with the image labelling.

Our approach to AI projects

To ‘teach’ a potato sorting machine what is a potato and what isn’t, our remote team annotated thousands of photos of potatoes using four specific tags: potato, leaves, stones and the machine’s conveyor.

Together with our client, precise specifications and guidelines were drawn up for labelling the photos. These stated which tags are to be used and what criteria need to be adopted to guarantee consistent, high-quality labelling. The labelling itself used an advanced tool called Label Studio plus our client’s own techniques to apply accurate labelling to the correct objects in the images.

Our remote team follow the guidelines carefully and apply strict quality checks to ensure that the labelled photographs meet the client’s expectations.

“Because human eyes and hands are still always needed for providing relevant training data for the computer systems and assessing whether the output is correct, we started working with DataMondial. Manually labelling photos is a time-consuming and expensive task for our internal team, so the choice to bring in an experienced remote team was an easy one.”

Do you also need help with an AI project?

If you’d like to know what we could do for your company, please contact us to make an appointment. Take a look at our page about data validation for OCR, AI and machine learning for more information too.