Panoptic One-Click Segmentation: Applied to Agricultural Data

In weed control, precision agriculture can help to greatly reduce the use of herbicides, resulting in both economical and ecological benefits. A key element is ability locate segment all plants from image data. Modern instance segmentation techniques achieve this, however, training such systems requires large amounts hand-labelled data which expensive laborious obtain. Weakly supervised labelling efforts costs. We propose panoptic one-click segmentation, an efficient accurate offline tool produce pseudo-labels click inputs reduces effort. Our approach jointly estimates pixel-wise location N objects scene, compared traditional approaches iterate independently through objects; this time. Using just 10% train our yields 68.1% 68.8% mean object intersection over union (IoU) on challenging sugar beet corn respectively, providing comparable performance while being approximately 12 times faster train. demonstrate applicability system by generating clicks remaining 90% These are then used Mask R-CNN, a semi-supervised manner, improving absolute (of foreground IoU) 9.4 7.9 points for respectively. Finally, we show that technique recover missed during annotation outlining further benefit approaches.