@misc{indiciaeaba26013e2ff,
  title = {Propose and Attend: Training-free MLLM Grounding Confidence via Multi-Token Localized Attention},
  author = {Daniel Shalam and Emanuel Ben Baruch and Avi Ben Cohen and Tal Remez},
  year = {2026},
  url = {https://arxiv.org/abs/2607.05978},
  note = {Source identifier: 2607.05978}
}