PixelGen

Low-power embedded cameras for mixed reality, with a generative model standing in for the image sensor.

PixelGen
ACM MobiCom '24 ImmerCom WorkshopACM/IEEE IPSN '24 Demo

Conventional embedded cameras capture only visible light and spend most of their energy on the image sensor and on streaming pixels. PixelGen instead senses the environment with a handful of low-power sensors, including audio, magnetic and other modalities, and transmits that compact description at a fraction of the energy. A multimodal generative model on the receiving side reconstructs a rich visual representation, which can be viewed on a mixed-reality headset.

PixelGen was published at the ImmerCom workshop at MobiCom ‘24, and its demonstration received the Best Demo Runner-Up award at IPSN ‘24.

Publications

  1. pixelgen.png
    PixelGen: Rethinking Embedded Cameras for Mixed-Reality
    Kunjun Li*, Manoj Gulati*, Dhairya Shah*, Steven Waskito, Shantanu Chakrabarty, and Ambuj Varshney
    ACM ImmerCom (MobiCom) ’24 Workshop
  1. pixelgen.png
    Demo Abstract: PixelGen: Rethinking Embedded Camera Systems for Mixed-Reality
    Kunjun Li*, Manoj Gulati*, Dhairya Shah, Steven Waskito, Shantanu Chakrabarty, and Ambuj Varshney
    ACM/IEEE IPSN ’24 Demo 🏅 Best Demo Runner-Up Award

All projects