search

BradyFU / Woodpecker

✨✨Woodpecker: Hallucination Correction for Multimodal Large Language Models. The first work to correct hallucinations in MLLMs.
580 stars 29 forks source link

How to reproduce results on POPE? #11

Closed kai-wen-yang closed 7 months ago

kai-wen-yang commented 7 months ago

Could you please provide the script that can reproduce results on POPE?

BradyFU commented 7 months ago

Hi Kaiwen, You can use our model to get the outputs, and it would be not difficult for you to calculate the indicators on POPE. Please feel free to contact me if you meet any questions.

kai-wen-yang commented 7 months ago

I don't know the hyperparameter (e.g. temperature) and prompt used when evaluating on POPE.

xjtupanda commented 7 months ago

I don't know the hyperparameter (e.g. temperature) and prompt used when evaluating on POPE.

Thanks for your interest in our work. We used default hyperparameters for the whole setup, except that the img_num is set to 50 for POPE. Feel free to ask here or through other means.