Fig. 3
From: PhenoTrack3D: an automatic high-throughput phenotyping pipeline to track maize organs over time
Deep-learning-based collar detection for organ segmentation. A Identification of the stem path (orange line) on the 3D skeleton, B reprojection of the stem path (orange line) on one of the corresponding RGB images and extraction of 416 × 416 sub-images along the stem path. Sub-images centers are evenly spaced along the stem path, with a maximum spacing of 400 pixels. C Collar detection on a sub-image (square: predicted bounding box, point: centre of the box, value: prediction score) D Projection of the highest detected collars point among all sub-images from all 12 side viewpoints on the 3D skeleton (grey cross). This point is then used to segment skeleton branches in stem, ligulated leaves, and growing leaves