Applications images and Input files for Deep Reinforcement Learning methods for AI task offloading in Smart Eyewears

This dataset contains input configuration files for a deep reinforcement learning approach to optimize the offloading of Deep Neural Network (DNN) computations from Smart EyeWear (SEW) devices to smartphone and cloud server. Project Overview The system can use different DRL methods (DQN, DDQN, Rainbow, PPO, A2C, among others) to learn optimal offloading policies that balance multiple competing objectives: Minimizing energy consumption on resource-constrained devices Meeting latency constraints for real-time applications Adapting to dynamic network conditions Dataset Description The dataset covers three AI tasks, each using a different DNN model: Object detection — YOLOv5n Pose estimation — YOLOv8n-Pose Classification — MobileNetV2 Each task is provided as a .zip archive containing 5 applications (app1 to app5) to ensure the robustness of the approach. Each application folder contains 3 files: system.json, initial_config.json, and env_config.json. A shared traces.zip archive provides network trace files used across all applications. More details regarding each file are provided in the readme.md file.