1 paper · 1 filter
Arjun Somayazulu, Efi Mavroudi, Changan Chen +2
Traditional methods for view-invariant learning rely on controlled multi-view training data with minimal scene clutter. However, they struggle with in-the-wild videos that exhibit…