Search papers, labs, and topics across Lattice.
Xi'an Jiaotong University
6
0
8
Treating negative samples differently based on their similarity to positives leads to a 13.1% boost in retrieval performance on complex queries.
CamFlow+ achieves superior 2D camera motion estimation and stabilization by seamlessly integrating multiple motion bases, outperforming traditional methods in complex environments.
DrPO achieves superior alignment in one-step generative models while slashing training computation costs by over 3x, challenging the status quo of preference finetuning.
Fix freaky AI-generated hands and other anatomical nightmares with a new preference learning method that surgically corrects anatomical errors in diffusion models.
Self-improving navigation agents can get lost in their own exploration unless you carefully balance behavioral diversity with learning stability, as shown by a new method that outperforms prior VLN models.
Instruction understanding can evolve dynamically with context, leading to a notable +2.68% SPL improvement in navigation tasks.
