Dear all,
you are cordially invited to join the master thesis defense of Gülsah Bauer with the topic given below. It will start at 9:30 on the 14th of January, 2022. We will use the normal virtual colloquium zoom room that you can find in this vc course.
Looking forward to it
Johannes
Topic: Effect of Hyperparameter Optimization on the Semantic Segmentation of Crops (MA CitH, with Fraunhofer IIS/EZRT Oliver Scholz)
Abstract:
This thesis presents the Dynamic Spherical Sub-Sampling (DSSS) algorithm, which operates on raw point cloud data. The DSSS prepares the 3D point clouds for 3D deep learning frameworks by generating spherical segments of the input data points in a user-defined size. These segments, named sub-spheres, are utilized as input for the 3D deep learning algorithms. The parameters in the generation process of the sub-spheres affect the learning results. PointNet++ and Dynamic Graph Convolutional Neural Networks (DGCNN) architectures are trained with two different data sets to demonstrate the effect of the hyperparameter optimizations on the semantic segmentation of the point clouds. In addition to the Wheat data set from Rothamsted Research and Fraunhofer IIS, the Forest data set from 3D Forest Project are employed as input data. The results show that an optimized relation between the hyperparameters improves the semantic segmentation performance.