search

bytedance / tarsier

Apache License 2.0
133 stars 8 forks source link

Prompts used in models evaluation on DREAM-1k #8

Open Dima-HV opened 3 weeks ago

Dima-HV commented 3 weeks ago

Hi, Great work!

Did you use the same prompt for all models evaluated on DREAM-1k? If not, what prompts did you use for different models?

jwwang424 commented 6 days ago

For all models, we use the default prompt "Describe the video in detail." unless the official implementation recommends a specific one. See Appendix C in the paper for the specific prompts that we have used.