Abstract: While pre-trained large-scale vision models have shown significant promise for semantic correspondence, their features often struggle to grasp the geometry and orientation of instances. This ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results