Search papers, labs, and topics across Lattice.
3
0
4
TTSD-FAR maintains consistent performance in emotion recognition even when up to half of the input modalities are missing, outperforming traditional adaptation methods.
Training with OpenMedReason boosts VQA accuracy by 20%, showcasing a leap in reasoning capabilities for medical vision-language models.
Fine-grained preference optimization can dramatically enhance the clinical accuracy of LVLMs by correcting only the clinically erroneous spans while preserving linguistic style.
