Search papers, labs, and topics across Lattice.
3
0
6
0
Grounding motion predictions in 3D spatial context and language leads to significantly more accurate and coherent forecasts of human actions.
FrozenDrive generates high-fidelity driving scenes under adverse conditions without sacrificing pre-trained model knowledge, outperforming existing methods.
Achieve state-of-the-art 4-bit LLM quantization accuracy with SERQ, a saliency-aware error reconstruction method that uses a single low-rank matrix, outperforming existing methods while reducing calibration complexity.
