search

TheShadow29 / VidSitu

[CVPR21] Visual Semantic Role Labeling for Video Understanding (https://arxiv.org/abs/2104.00990)
https://vidsitu.org/
MIT License
56 stars 8 forks source link

Verbs missing? #6

Open brandonclark314 opened 3 years ago

brandonclark314 commented 3 years ago

Hello,

I used your verb_id_vocab.pkl to get the numeric labels for the verbs, but I noticed that there are some verbs that have been assigned to videos in the validation split that are not in verb_id_vocab.pkl. I also noticed that the pickle file only includes about 1.5k verbs but your paper mentions having around 2.1k verbs. Is it possible I'm not getting the labels from the right place, or maybe some issues with my handling of the .pkl file?

Thanks

TheShadow29 commented 3 years ago

@brandonclark314 Thanks for your interest!

2.1k is the number of verbs available to the annotators. But not all were used.

In the training set, 1560 verbs in total are used, which is reflected in the pkl file. Does that clarify your question?