Search papers, labs, and topics across Lattice.
University of Ljubljana
3
0
5
SIFT features can outperform some of the latest deep-learning image matching methods in UAV visual odometry, challenging the assumption that newer always means better.
Multispectral Earth observation models can now learn more effectively from optical vision foundation models using a dual-teacher distillation approach, boosting performance by up to 3.64% on key tasks.
By synthesizing changes directly in the latent space, MaSoN achieves state-of-the-art unsupervised change detection in remote sensing, generalizing to complex and rare scenarios where existing methods fall short.
