Enhancing performance in video grounding tasks through the use of attention module
This report investigates improving video grounding tasks through the use of attention mechanisms, tackling the issue of sparse annotations in video datasets. Drawing inspiration from the MMN model \cite{wang2021_negative_2dmap}, we developed a modified model based on the open-source MMN codebase and...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
Nanyang Technological University
2024
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/181703 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | This report investigates improving video grounding tasks through the use of attention mechanisms, tackling the issue of sparse annotations in video datasets. Drawing inspiration from the MMN model \cite{wang2021_negative_2dmap}, we developed a modified model based on the open-source MMN codebase and evaluated it on several widely-used datasets, including Charades-STA and ActivityNet Captions. Our approach shows improvements over certain benchmarks. Additionally, we conducted an in-depth analysis to assess the role of attention in enhancing the multimodal framework's ability to comprehend the complex structure of videos. |
---|