saccharomycetes / text-based-traffic-understanding

The code and dataset of the KDD23 paper 'A Study of Situational Reasoning for Traffic Understanding'
GNU General Public License v3.0
15 stars 1 forks source link

Dataset #1

Closed monjurulkarim closed 6 months ago

monjurulkarim commented 6 months ago

How to get BDD-QA and HDT-QA Dataset?

saccharomycetes commented 6 months ago

Hi Muhammad, the two data are under the /data folder of our repo, bdd.jsonl and hdt.jsonl

monjurulkarim commented 6 months ago

@saccharomycetes Thanks for your response. I found the annotation. Could you please direct me to where I can download the corresponding videos?

saccharomycetes commented 6 months ago

You are welcome, here is the source of the BDD-QA dataset (BDD-X). But to my knowledge, their release of videos is lost. Since our research focuses on textual data only, we did not use the videos of BDD-X. HDT-QA crawled from here, they are text-based traffic-related questions so there is no corresponding video. For our TV-QA dataset, you can download the corresponding videos here, what we did is we provided detailed textual annotations (captions) for each video (a subset of videos in traffic QA) and checked the quality of every question and caption manually and carefully. So I think TV-QA is a good testbed if you want to explore topics about video-QA, causality, etc.

monjurulkarim commented 6 months ago

@saccharomycetes Thanks for your detailed response. I agree that TV-QA appears to be an excellent match for our needs. I reached out to here last week to request access to the dataset but haven't received a response yet. Hopefully, they will provide a download link soon.

saccharomycetes commented 6 months ago

Of course, I think if you submit your information here, you can immediately get the link to the data through Google Drive.

monjurulkarim commented 6 months ago

@saccharomycetes Excellent! Many thanks!