PixelCraftSR: Efficient Super-Resolution with Multi-Agent Reinforcement for Edge Devices.

Area of Science:

• Computer Vision
• Artificial Intelligence

Background:

• Super-resolution (SR) imaging is vital across many fields, including medical imaging and digital displays.
• Current deep learning-based SR methods are computationally intensive, limiting their use on resource-constrained Internet of Things (IoT) devices.

Purpose of the Study:

• To propose a lightweight and efficient SR model suitable for IoT applications.
• To leverage multi-agent reinforcement learning for improved image reconstruction.

Main Methods:

• A novel lightweight model employing a multi-agent reinforcement learning (MARL) approach.
• Pixel-level agents utilizing an asynchronous actor-critic policy to construct SR images.
• Iterative action selection based on image state to maximize cumulative reward over five time steps.

Main Results:

• The proposed MARL-based SR method outperforms existing SR techniques in qualitative and quantitative evaluations.
• Achieves significantly lower computational complexity compared to current state-of-the-art methods.
• Demonstrates practical applicability on various IoT platforms, including edge devices.

Conclusions:

• The developed lightweight MARL model offers an efficient solution for single-image super-resolution on IoT devices.
• Provides a viable alternative to computationally expensive SR methods for edge computing scenarios.