Gaze dataset coverage (v1 Vs. v2)

Q&A
, , ,
1

Hi colleagues,

I am currently utilizing the online visualizer to examine the statistics of our dataset, with a particular focus on gaze datasets. I have some inquiries regarding the correlation between version 1 and version 2.

Inputting filter modalities include gaze, here are the results,

  1. Version 1
    Browsing 2293 / 9645 videos . Total Duration: 866.21 hours .
  2. Version 2
    Browsing 31 / 9611 videos . Total Duration: 33.35 hours .

In my understanding, this means,

  1. In v1, a total of 9645 - 9611 = 34 videos were removed due to the reason outlined here.
  2. In v2, there are an additional 31 gaze data published, which means the corresponding videos are present in v1 but their gaze data were not published in v1.
  3. If we want to use as much gaze data as possible, ideally, there are 2293 + 31 = 2324 videos that are eligible/valid to use.

Please kindly correct me if something goes wrong. Thank you!

2

We do not have 2293 videos with Gaze for V1. There is only 31 videos. This is a bug in the visualizer’s metadata or the V1 metadata. Please use V2 as a source of truth.

Sorry for the inconvenience