Dense Dilated Network for Video Action Recognition.

Area of Science:

• Computer Vision
• Machine Learning
• Artificial Intelligence

Background:

• Video action recognition is crucial for surveillance and autonomous driving.
• Deep neural networks for action recognition typically require extensive labeled data.
• Existing methods face challenges in achieving high performance with limited training data.

Purpose of the Study:

• To develop a novel deep neural network framework for efficient video action recognition.
• To address the challenge of limited labeled data in training action recognition models.
• To improve the robustness and performance of video action recognition systems.

Main Methods:

• Introduction of a dense dilated network architecture.
• Utilizing densely connected dilated convolutions for feature extraction.
• Implementing a novel temporal guided fusion mechanism for enhanced representation learning.

Main Results:

• The proposed dense dilated network effectively captures action information from snippet to global levels.
• The framework demonstrates robust performance even with few training snippets.
• Extensive experiments on UCF101 and HMDB51 datasets validate the framework's effectiveness.

Conclusions:

• The dense dilated network offers an effective solution for video action recognition with limited data.
• The temporal guided fusion significantly boosts recognition performance.
• The proposed framework shows strong potential for real-world applications like surveillance and autonomous driving.