PhD student, Department of Computer Science, University of Wisconsin - Madison
2 papers at NeurIPS 2025
We propose GLSim, a training-free framework that combines global and local embedding similarity signals for accurate object hallucination detection in LVLMs, outperforming prior methods.
