Published on:
22 April 2024
Primary Category:
Networking and Internet Architecture
Paper Authors:
Mathias Thorsager,
Victor Croisfelt,
Junya Shiraishi,
Petar Popovski
Uses tiny ML models on sensors for energy-efficient image retrieval
Behavior model filters irrelevant images before transmission
Image compressor reduces bandwidth via latent representations
Saves 70%+ energy vs uncompressed transmission
Maintains image quality with compression
Energy-efficient image retrieval from wireless sensors
This paper proposes EcoPull, an energy-efficient framework for image retrieval from wireless visual sensors. It uses two tiny machine learning models on the sensors: a behavior model to filter out irrelevant images, and an image compressor to reduce transmission bandwidth. Together these models save over 70% energy compared to uncompressed image transmission, while maintaining image quality.
Energy-efficient image retrieval using tiny ML models
Harnessing mobile crowdsensing and wireless power transfer for sustainable and privacy-preserving...
Efficient image transmission through neural networks
Efficient image communication for AIoT using deep semantic segmentation and restoration
Identifying transmitters through image and signal data
Trading energy for output quality in battery-less devices
No comments yet, be the first to start the conversation...
Sign up to comment on this paper