Georgios Balaouras

Research

Publications

Three papers on supervised and unsupervised video summarization, and on explaining the resulting summaries.

My complete publication record on Google Scholar

ISM2022 2022

Explaining video summarization based on the focus of attention

Shows which video fragments influence a summary using signals from the model's attention mechanism.

IEEE Int. Symposium on Multimedia (ISM) 2022 December 2022

Evlampios Apostolidis*, Georgios Balaouras*, Vasileios Mezaris, Ioannis Patras

* Equal contribution

PDF (opens in a new tab) DOI (opens in a new tab) Code (opens in a new tab)
View citations Google Scholar (Citation count from Semantic Scholar. Link opens Google Scholar.)
Diagram showing how video frames pass through an attention network to produce a summary and an explanation mask.
Overview of the attention-based explanation method.

ICMR2022 2022

Summarizing Videos using Concentrated Attention and Considering the Uniqueness and Diversity of the Video Frames

Uses concentrated attention and frame uniqueness to create video summaries without supervision.

ACM Int. Conference on Multimedia Retrieval 2022 June 2022

Evlampios Apostolidis*, Georgios Balaouras*, Vasileios Mezaris, Ioannis Patras

* Equal contribution

PDF (opens in a new tab) DOI (opens in a new tab) Code (opens in a new tab) Video (opens in a new tab) Slides (opens in a new tab)
View citations Google Scholar (Citation count from Semantic Scholar. Link opens Google Scholar.)
CA-SUM architecture: a CNN and concentrated attention estimate the importance of video frames.
Architecture of the concentrated-attention summarization model.

ISM2021 2021

Combining Global and Local Attention with Positional Encoding for Video Summarization

Combines global and local attention with positional encoding for supervised video summarization.

IEEE Int. Symposium on Multimedia (ISM) 2021 December 2021

Evlampios Apostolidis*, Georgios Balaouras*, Vasileios Mezaris, Ioannis Patras

* Equal contribution

PDF (opens in a new tab) DOI (opens in a new tab) Code (opens in a new tab) Video (opens in a new tab) Slides (opens in a new tab) Project (opens in a new tab)
View citations Google Scholar (Citation count from Semantic Scholar. Link opens Google Scholar.)
PGL-SUM architecture combining global and local multi-head attention to score video frames.
Architecture combining global and local attention.