Learning Recurrent Memory Activation Networks for Visual Tracking.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Deep neural networks have advanced visual tracking.
• Current methods often overlook target appearance's temporal coherence, relying on independent frames.
• This limitation can lead to reduced tracking accuracy and increased background interference.

Purpose of the Study:

• To propose a novel Recurrent Memory Activation Network (RMAN) for visual tracking.
• To exploit the temporal coherence of target appearance for improved tracking performance.
• To develop a fully differentiable network optimized end-to-end.

Main Methods:

• The RMAN is built upon the Long Short-Term Memory (LSTM) network with an added memory activation layer.
• LSTM models temporal changes in target appearance.
• The memory activation layer selectively activates memory blocks for temporally coherent representations.

Main Results:

• The RMAN effectively enriches target representations by integrating temporal information.
• Temporal consistency reduces background interference.
• Experimental results on standard benchmarks show favorable performance against state-of-the-art methods.

Conclusions:

• The proposed RMAN enhances visual tracking by effectively utilizing temporal coherence.
• The integration of a temporal coherence loss facilitates network training.
• RMAN offers a promising approach for robust and accurate visual tracking.