Maciej Wielgosz
Research Scientist
Biography
My expertise lies in the field of machine learning; I'm dealing with Point Cloud segmentation models. My primary focus is to address the challenges posed by sparse sections of point clouds, especially those that are crucial for forestry applications, like the sections near the base of tree trunks. While data from drones and airplanes is readily accessible, ensuring high semantic accuracy during processing can be quite intricate. Therefore, new methods in point cloud instance and semantic segmentation are needed.
Abstract
No abstract has been registered
Abstract
Accurately predicting whether pedestrians will cross in front of an autonomous vehicle is essential for ensuring safe and comfortable maneuvers. However, developing models for this task remains challenging due to the limited availability of diverse datasets containing both crossing (C) and non-crossing (NC) scenarios. Therefore, we propose a procedure that leverages synthetic videos with C/NC labels and an untrained model whose architecture is designed for C/NC prediction to automatically produce C/NC labels for a set of real-world videos. Thus, this procedure performs a synth-to-real unsupervised domain adaptation for C/NC prediction, so we term it S2R-UDA-CP. To assess the effectiveness of S2R-UDA-CP in self-labeling, we utilize two state-of-the-art models, PedGNN and ST-CrossingPose, and we rely on the publicly-available PedSynth dataset, which consists of synthetic videos with C/NC labels. Notably, once the real-world videos are self-labeled, they can be used to train models different from those used in S2R-UDA-CP. These models are designed to operate onboard a vehicle, whereas S2R-UDA-CP is an offline procedure. To evaluate the quality of the C/NC labels generated by S2R-UDA-CP, we also employ PedGraph+ (another literature referent) as it is not used in S2R-UDA-CP. Overall, the results show that training models to predict C/NC using videos labeled by S2R-UDA-CP achieves performance even better than models trained on human-labeled data. Our study also highlights different discrepancies between automatic and human labeling. To the best of our knowledge, this is the first study to evaluate synth-to-real self-labeling for C/NC prediction.
Authors
Maria Åsnes Moan Stefano Puliti Rasmus Astrup Ole Martin Bollandsås Terje Gobakken Maciej Wielgosz Hans Ole Ørka Lennart NoordermeerAbstract
Abstract The site index (SI) describes a site’s potential to produce wood volume. Accurate information on SI in young forests is essential for planning thinning operations and projecting future growth and yield. For tree species that form annual branch whorls, information on interwhorl distances along the stem may be used to determine the SI in young forests. Branch whorls, and consequently tree height growth trajectories, can be detected automatically using deep learning on very dense laser scanning data. In the current study, we demonstrate this approach in a case study in a young Norway spruce forest. We trained a pose estimation Convolutional Neural Network and detected branch whorls of 97 dominant trees in 54 plots scanned with mobile laser scanning data. We predicted SI determined from detected branch whorls in three different sections of each tree, selected in the stem height range between 2.5 and 8 m: all whorls, the lowest six whorls, and whorls selected with an automatic selection procedure. We compared the obtained SI to the SI determined from field-measured branch whorls. Obtained values of precision, recall, and F1 score for the branch whorl detection were 0.66, 0.58, and 0.62, respectively. Values of root mean square error and mean differences between reference and predicted SI ranged between 19.8%–20.9% and −3.6%–4.0%, respectively. Although the tested approach showed potential for SI determination in young forests, the obtained errors were large. This was due to detection errors and high sensitivity to small changes in height increment. These issues highlight the need for further research to improve branch whorl detection accuracy and address challenges associated with determining the SI in young forests.
