Table 2 in paper

[noor-fathima-khanum]

Hello,

Thank you for releasing I2V evaluation benchmark. It is super useful for the community.

I am trying to replicate scores of Table-2 in your paper.
I had some questions:

1. Which dataset was used to evaluate these I2V models?
2. How did you extract the first frame to feed as conditioning?
3. Were the 1000 WebVid10M validation videos used as Reference videos?

Your inputs would be very helpful!
Thank you,
Noor

[fanfanda]

Hello, thanks for your attention.

1. Please refer to the https://huggingface.co/datasets/stevenfan/AIGCBench_v1.0.
2. We extract the 10th frame in WebVid10M dataset.
3. Yes, we used webvid list in https://huggingface.co/datasets/stevenfan/AIGCBench_v1.0.