MMVD

Multimodal Video Description (MMVD) framework generates rich natural langauge description of web videos by exploring information from multiple sources e.g., viasual concepts, audio tracks and video category information. The framework is one of the top three performers in the MST-VTT grand challenge organized with ACM-MM 2016. This page provides the code, additional results and illustrative examples of the challenges involved in producing natural language description for "in the wild" web videos as found in MSR-VTT dataset.