ROBUST SPARSE AND DENSE MATCHING
Primary Area: representation learning for computer vision, audio, language, and other modalities
Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.
Keywords: sparse matching, dense matching, optical flow, geometry estimation
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Finding corresponding pixels within a pair of images is a fundamental computer vision task with various applications. Due to the specific requirements of different tasks like optical flow estimation and local feature matching, previous works are primarily categorized into dense matching and sparse feature matching focusing on specialized architectures along with task-specific datasets, which may somewhat hinder the generalization performance of specialized models. In this paper, we propose RSDM, a robust network for sparse and dense matching. A cascaded GRU module is elaborately designed for refinement to explore the geometric similarity iteratively at multiple scales following an independent uncertainty estimation module for sparsification. To narrow the gap between synthetic samples and real-world scenarios, we organize a new dataset with sparse correspondence ground truth by generating optical flow supervision with greater intervals. In due course, we are able to mix up various dense and sparse matching datasets significantly improving the training diversity. The generalization capacity of our proposed RSDM is greatly enhanced by learning the matching and uncertainty estimation in a two-stage manner on the mixed data. Superior performance is achieved for zero-shot matching as well as downstream geometry estimation across multiple datasets, outperforming the previous methods by a large margin.
Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.
No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.
Submission Number: 2298
Loading