How to generate the LOL dataset ?

[TriLoo]

Hi there, I wonder how can I generate the LOL dataset using my own feature extractor ?

Now I want to use the trained network to generate summary on custom videos, but I don't know the exact feature extractor (network) so the extracted feature vectors on TVSum is different from that from downloaded .h5 files. This could lead to a performance drop because of the input data distribution mismatch.

I have tried the GoogleNet pool5 from Torchvision pretrained model zoo, but look like the values are different totally.

Thanks for any advices.