大幅超越SuperGlue！ICCV2023最新端到端特征匹配方法
TIP! Right-click and select "Save link as..." to download.
| VIDEOS | |||
| MP4 | N/A | 480P | Download |
| MP4 | N/A | 480P | Download |
| MP4 | N/A | 360P | Download |
| MP4 | N/A | 360P | Download |
| AUDIO | |||
| MP4 | N/A | mp4a.40.2 | Download |
| MP4 | N/A | mp4a.40.5 | Download |
| MP4 | N/A | mp4a.40.2 | Download |
| THUMBNAILS | |||
|
JPEG | Origin Image | Download |
错误的特征匹配对后续的相机姿态估计有严重影响，并且通常需要额外的、耗时的措施，如RANSAC，来拒绝异常值。我们的方法通过联合处理特征匹配和姿态优化来应对这一挑战。与SuperGlue相比，多视图匹配与端到端训练相结合将Matterport3D的姿势估计指标提高了18.5%。
论文链接：https://arxiv.org/abs/2205.01694
https://www.youtube.com/watch?v=uuLb6GfM9Cg
论文链接：https://arxiv.org/abs/2205.01694
https://www.youtube.com/watch?v=uuLb6GfM9Cg
Supported Sites
