Computer Engineering / Bilgisayar Mühendisliği
Permanent URI for this collectionhttps://hdl.handle.net/11147/10
Browse
2 results
Search Results
Article Label-Free Retraining for Improved Ground Plane Segmentation(Springer, 2022) Uzyıldırım, Furkan Eren; Özuysal, MustafaDue to increased potential applications of unmanned aerial vehicles over urban areas, algorithms for the safe landing of these devices have become more critical. One way to ensure a safe landing is to locate the ground plane regions of images captured by the device camera that are free of obstacles by deep semantic segmentation networks. In this paper, we study the performance of semantic segmentation networks trained for this purpose at a particular altitude and location. We show that a variation in altitude and location significantly decreases network performance. We then propose an approach to retrain the network using only a new set of images and without marking the ground regions in this novel training set. Our experiments show that we can convert a network’s operating range from low to high altitudes and vice versa by label-free retraining.Article Citation - WoS: 5Citation - Scopus: 5Instance Detection by Keypoint Matching Beyond the Nearest Neighbor(Springer Verlag, 2016) Uzyıldırım, Furkan Eren; Özuysal, MustafaThe binary descriptors are the representation of choice for real-time keypoint matching. However, they suffer from reduced matching rates due to their discrete nature. We propose an approach that can augment their performance by searching in the top K near neighbor matches instead of just the single nearest neighbor one. To pick the correct match out of the K near neighbors, we exploit statistics of descriptor variations collected for each keypoint in an off-line training phase. This is a similar approach to those that learn a patch specific keypoint representation. Unlike these approaches, we only use a keypoint specific score to rank the list of K near neighbors. Since this list can be efficiently computed with approximate nearest neighbor algorithms, our approach scales well to large descriptor sets.