Questions about Figure 3 in paper

shikiw / OPERA

[CVPR 2024 Highlight] OPERA: Alleviating Hallucination in Multi-Modal Large Language Models via Over-Trust Penalty and Retrospection-Allocation

MIT License

244 stars 22 forks source link

Questions about Figure 3 in paper #18

Closed minhoooo1 closed 3 months ago

minhoooo1 commented 4 months ago

We are interested in Figure 3 in the paper. Because it seems to be the starting point of the paper. We would like to know how you define knowledge aggregation patterns when collecting data from Figure 3. Is it through the naked eye? And how was the "within 10 tokens" calculated?

Looking forward to your reply, thank you！

shikiw commented 4 months ago

Hi,

Thanks for your interest! Yes, we collect the data in Figure 3 and calculate the results of "within 10 tokens" by naked eyes. You can consider the visualization script vis.ipynb and https://github.com/shikiw/OPERA/issues/14 to visualize it for observation.