3DG-STFM: 3D Geometric Guided Student-Teacher Feature Matching

We tackle the essential task of finding dense visual correspondences between a pair images. This is challenging problem due to various factors such as poor texture, repetitive patterns, illumination variation, and motion blur in practical scenarios. In contrast methods that use correspondence ground-truths direct supervision for local feature matching training, we train 3DG-STFM: multi-modal model (Teacher) enforce depth consistency under 3D transfer knowledge 2D unimodal (Student). Both teacher student models consist two transformer-based modules obtain coarse-to-fine manner. The guides learn RGB-induced information purpose on both coarse fine branches. also evaluate 3DG-STFM compression task. To best our knowledge, first student-teacher learning method experiments show outperforms state-of-the-art indoor outdoor camera pose estimations, homography estimation problems. Code available at: https://github.com/Ryan-prime/3DG-STFM .