Search papers, labs, and topics across Lattice.
1
0
2
Pre-training objectives shape how ViTs encode geometric information, revealing that some models can compress dense geometric signals without sacrificing performance.
