SRSP: Sub-Random SuperPoint Based on Reprojection Error and Randomized Round Encoding
Xiaoyu Tian,Li Li,Hongyu Cao
DOI: https://doi.org/10.1109/access.2024.3443093
IF: 3.9
2024-08-21
IEEE Access
Abstract:Interest point detection and matching play an extremely important role in many computer vision applications. In recent years, deep learning-based algorithms are being used for interest point detection, of which SuperPoint is the most notable algorithm. Although SuperPoint has achieved good results, its interest point detection accuracy is limited to the pixel level; additionally, the simple rounding of noninteger ground truth coordinates in this algorithm results in the loss of decimal information, introducing quantization errors. To overcome these limitations, this study introduces the subpixel module and randomized round encoding methods to reduce the prediction error and quantization error in SuperPoint, respectively. First, we propose a differentiable decoder, soft-random to achieve subpixel-level accuracy of interest point detection. Additionally, to further reduce the interest point localization error, we propose the reprojection homography adaptation of the training step based on SuperPoint's homography adaptation that is, weincorporate the reprojection error into the algorithm's training loss. The optimized algorithm proposed in this study, called SRSP, is tested on the HPatches dataset and Euroc dataset. The results showed that SRSP performed better than SuperPoint regarding all indicators, indicating the effectiveness of SRSP on both image matching dataset and real-world dataset.
computer science, information systems,telecommunications,engineering, electrical & electronic