Search papers, labs, and topics across Lattice.
Affiliation:
5
0
6
Instead of forcing entire videos into a single, lossy feed-forward context, training an agent to actively read and write temporal memory yields 78.8% accuracy on MINERVA with virtually zero performance drop on long-duration clips.
Action-class miscalibration is a hidden failure mode that can significantly skew performance evaluations of LLMs in multi-turn tool-calling tasks.
Dynamo achieves a remarkable 5.6% average accuracy boost in visual reasoning tasks without retraining, revolutionizing how VLMs adapt in real-time.
VLMs and VGMs reveal a surprising complementarity in spatial intelligence tasks, with a simple fusion of their features outperforming either model alone.
MLLMs still fail at complex visual workflows: even the best models struggle to navigate multi-step reasoning chains involving visual conditions, achieving only 53.33 Path F1 on the new MM-CondChain benchmark.
