Large Scale Movie Description Challenge (LSMDC) 2019

Tasks

This year we introduce a new challenge, aiming at a more realistic and practical setting of multi-sentence movie description generation. Specifically, movie descriptions are evaluated on sets of 5 clips, rather than on individual clips. When describing sequences of events, it becomes important to distinguish "who is who" in order to provide a coherent and informative narrative. Thus, the challenge will have a focus on identifying characters, rather than predicting generic "SOMEONE"-s in place of all the occurring character names. The challenge is split in three tasks:

  • the 1st task still asks to generate descriptions with "SOMEONE"-s but the evaluation is carried out on sets of 5 clips,
  • the 2nd task focuses on filling-in the character IDs locally (within a set of 5 clips), while the rest of each description is available,
  • the 3rd task combines both description generation and filling-in the local character IDs for the sets of 5 clips.

More details are provided here and the corresponding annotations are available for download here.

Important dates

Challenge submission deadlines

Task 1 "Multi-Sentence Description": October 1st

Task 2 "Fill-in the Characters": October 1st

Task 3 "Multi-Sentence Description with Characters": October 1st

Workshop: Monday, Oct 28th (full day)

Contact

lsmdc2015+19 at gmail.com

Organizers

Anna Rohrbach, Research Scientist at UC Berkeley

Jae Sung Park, Student Assistant at UC Berkeley, PhD student at University of Washington