Search papers, labs, and topics across Lattice.
3
0
5
EQA agents can now handle dynamic, human-populated scenes better thanks to a training-free method that selectively remembers only the most informative visual evidence.
A million-scale dataset of globally diverse, cross-modal geo-location pairs, coupled with a novel physical-law-aware network, leapfrogs existing CMGL benchmarks and opens the door to truly universal positioning systems.
Forget training separate models for every remote sensing modality pair: Any2Any learns a single latent space for unified translation, even generalizing to unseen modality combinations.
