A tool lets farm robots label plants and fruits with just one click, sometimes even none.

Farming robots need to see and outline objects like crops and fruit, a task called segmentationsegmentationLabeling each pixel in an image to identify specific objects like plants.. But humans must hand-label thousands of pictures to teach them, which is slow and expensive. To fix this, researchers created DropClick. It is a tool that needs only a single mouse click on an object to outline it automatically and make training labels. Best of all, users do not even need to click on every single item in the picture. DropClick learns to spot the rest on its own. The researchers trained the tool using just 5 images. Even when half of the user clicks were left out, the system kept working well. Using these computer-made labels to train another vision model gave nearly the same strong results as clicking every single item. This simple trick saved workers up to 46.3% of their clicking effort.

Why it matters

Labeling images for agricultural robots is slow and expensive, which delays better farm automation. This tool cuts the human work needed, helping robots learn to see crops faster.

Questions answered