TL;DR: We make better, cheaper night vision.

🌒 The Problem:

Seeing in near-total darkness is usually solved with specialized and expensive hardware like the $3000 image intensifier tube. Solutions like thermal and infrared cameras are extremely expensive and power-hungry. Digital cameras such as CMOS are still unable to perform in near-total darkness like moonless starlight.

💡 Our solution:

We leverage state-of-the-art digital CMOS sensors, high-performance mobile SOCs, and edge AI processing to build real-time on-edge night vision that enables devices to outperform the state of the art in night vision technology. Operators will be able to see in the dark at framerate at a fraction of the cost. Unlike the analog image intensifier tube, our solution is digital, enabling useful downstream features like streaming and object recognition. The core enabling technology is AI signal processing that deciphers the extremely limited signal from the CMOS sensor in near-total darkness.

Here is the impact of our AI low light signal processing:

Our solution also works at frame rate on low-power devices like a mobile phone:

🤝 Team:

Thomas and Lucas grew up in New Jersey together and stayed in touch. Thomas went to UC Berkeley and studied EECS before hopping from Lawrence Berkeley National Laboratory to Lyft to Google, working primarily on ML systems. Lucas went to Cal Poly SLO and specialized in computational photography, researching edge computer vision problems at Meta and the Google Pixel team.

🙏 Our Asks

We are seeking commercial use cases for night vision technology e.g. camera systems limited in low light. If you are interested in collaboration, please get in touch! Send an email to thomas@deepnight.ai or lucas@deepnight.ai.

We are hiring - if you are a research scientist experienced in computational imaging and find this work exciting, please reach out!

We are supported by the US Army’s Night Vision and Electronic Sensors Directorate and have received an LOI from Anduril Industries!