The goal of this challenge is to offer a fertile ground for designing pre-training techniques that facilitate a series of video understanding downstream tasks (e.g., video captioning and video categorization this year). Hence, two tracks (Pre-training for Video Captioning and Pre-training for Video Categorization) will be involved in this grand challenge. Meanwhile, to further motivate and challenge the multimedia community, we provide two large-scale video pre-training datasets, i.e., Auto-captions on GIF (ACTION) and the Weakly-Supervised dataset, for contestants to solve this challenging but emerging task in each track.
Particularly, in the first track, the contestants are asked to develop video captioning system based on ACTION dataset (as pre-training data) and the public MSR-VTT benchmark (as training data for downstream task). For the evaluation purpose, a contesting system is asked to produce at least one sentence of the test videos. The accuracy will be evaluated against human pre-generated sentence(s).
In the second track, the contestants are asked to develop video categorization system based on the Weakly-Supervised dataset (as pre-training data) and the provided Downstream dataset (as training data for downstream task). During evaluation stage, a contesting system is asked to predict the category of the test videos. The accuracy will be evaluated against human annotated categories.
