Summarizer is a Video Summarization framework for research. Most of the literature now focuses on Deep Learning models experimenting on a set of reference datasets. This repository gathers the key assets to ease this research into a single Python framework.
The four main components are:
- Centralized, preprocessed and documented datasets
- PyTorch implementation and bugfixes of the prominent models
- A robust set of evaluation metrics to evaluate them
This framework is dedicated at helping to design the next generation of Deep Learning models for Video Summarization.
SumMe | TVSum | LOL |
---|---|---|
![]() |
![]() |
![]() |
Datasets ground truth scores distribution per video
The architecture of Summarizer was inspired by K. Zhou et al. and J. Fajtl et al.. The preprocessed datasets were inspired by K. Zhang et al.. We thank them all for their leading contributions.