LSMDC v2 Challenge

The Large Scale Movie Description and Understanding Challenge (LSMDC) is based on the M-VAD and MPII-MD datasets. It comprises 200 movies with 120K corresponding sentence descriptions, generated by transcribing Descriptive Video Service (DVS) audio for the blind. These DVS descriptions are of high quality, and focused on relevant visual information in the scene, making them ideal for training video and natural language models. Over 400 research groups have accessed this dataset so far. The details on the challenge tracks present this year are available on the Challenge website.