the result is not as good as the result in demo

What are the parameters in the demo? Why can't I get the good result showed in the demo even though I used the same image? Does using different models make a big difference to the results?

Start by comparing the example notebooks and the model checkpoints used for the demo with the reported setup. Check the inference parameters and model choice against the repository's documented example workflow; done would require identifying a reproducible cause for the result difference and documenting the matching configuration.

Written by the indexing model from the issue text.