Learning visual overlapping image pairs for SfM via CNN fine-tuning with photogrammetric geometry information

Qianbao Hou, Rui Xia, Jiahuan Zhang, Yu Feng, Zongqian Zhan, Xin Wang

Abstract: Highlights•Faster image matching has been achieved while preserving the SfM result quality.•A fine-tuning method for various CNN models is developed via embedding local regional overlapping information.•A benchmark including both crowdsourced and photogrammetric images is published for testing and training overlapping image pair methods.•The benchmark is automatically generated with photogrammetric geometry information.