Search papers, labs, and topics across Lattice.
1
0
3
By forcing VLMs to treat 2D features as "what" and 3D features as "where," this method significantly boosts 3D grounding accuracy without modifying the model architecture.
