Awesome

PororoQA

Authors: Kyung-Min Kim, Min-Oh Heo, Seong-Ho Choi, and Byoung-Tak Zhang (Seoul National University & Surromind Robotics)

Paper1: DeepStory: Video Story QA by Deep Embedded Memory Networks (https://arxiv.org/abs/1707.00836) (IJCAI 2017)

Paper2: PororoQA: A Cartoon Video Series Dataset for Story Understanding (NIPS 2016 Workshop on Large Scale Computer Vision System, 2016)

Download Video Clips

You can download the whole Pororo video clips with this URL.
If this link doesn't work, please email me kyungmin.kim.ml@navercorp.com

Why cartoon video seires?

Characteristic of “Pororo”, children's cartoon video series

• Easy sentence
• A simple story structure
  • One episode: Avg. 7.2 min.
  • Similar events are repeated
• Small environments
  • The size of dictionary: 3,729
  • The number of characters: 13
• It may be appropriate as a test bed for video story QA problem

Data Statistics - QAs

• 8,834 QA pairs
• Remove 'ambiguous', 'unrelated' questions that do not follow guidelines
• Average QA count per episode : 51.66
• Examples and statistics by type of question
• Statistics about questions based on answer types
• Average number of words in our dataset depend on the first word in the question
• Comparison of various public datasets in terms of video story analysis

Data Statistics – Scene Descriptions

• 27,328 scene descriptions
  • Average sentence / word count per scene description : 1.7 / 13.6