Video Entity Linking Dataset

The dataset contains the dataset used by the paper [1] on video entity linking.

The README:

There are two batches, for disambiguration task and verification task, respectively

disam/verify

One match per subfolder. Each subfolder includes the following infomation,
1. keyframes/: the extracted keyframes for the video
2. images/: google search result for the entity
3. meta.json: video title, youtube url, entity and the label.

label = 1 means matched

`mil_ldml_struct.m` is the code for the proposed algorithm.

Reference:
Li, Yuncheng, Xitong Yang, and Jiebo Luo. "Semantic video entity linking based on visual content and metadata." ICCV 2015.
  

Reference

[1] Li, Yuncheng, Xitong Yang, and Jiebo Luo. "Semantic video entity linking based on visual content and metadata." ICCV 2015.