SDC – Stacked Dilated Convolutions

SDC – Stacked Dilated Convolutions
René Schuster, Oliver Wasenmüller, Christian Unger, Didier Stricker
Computer Science in Cars Sympoisum. ACM Computer Science in Cars Symposium (CSCS-2019) October 8-8 Germany ACM 10/2019 .

Abstract:

Dense matching is a fundamental problem in many tasks and applications of Computer Vision. Of utmost importance for robust matching algorithms is a powerful representation of image points. With SDC (Stacked Dilated Convolution), we have presented a universal design element that was successfully used in a deep neural network for dense feature description of images. Using these descriptors, we could improve matching in wide variety of problems and domains.