Fig. 1

Berry annotation procedure. A Raw labels, consisting of simple polygons (5 to 10 points) drawn manually along the edges of berries with at least 50% of their contour visible. B Guessed actual contour of berries, obtained by an automatic ellipse fitting (blue) on the annotated points. C Instances generated from the annotations dataset, used to train the segmentation model. Each instance corresponds to one berry, for which we show the fitted ellipse parameters, the image input and the targeted binary segmentation mask