Computer Vision Engineer

Deepnight is looking for a computer vision and deep learning engineer to advance the state of the art in low-light imaging. You will design novel neural architectures, train models, and deploy real-time inference on embedded hardware, working from your own expertise and from what you pull out of public research. The whole job comes down to one thing: the most accurate model that runs as fast as possible, inside a one cubic inch camera at 90 FPS on a single watt.

If you have had an idea killed for reasons that had nothing to do with whether it would work, you will feel at home here.

What You'll Do

Design and modify model architectures. Change the network, not the code around it, and take image quality and latency as the things you are judged on.

Run your own experiments. If you think a change makes the model better or faster, you build the test and you run it, without waiting for someone to scope it.

Deploy to constrained hardware. Take models to edge AI chips under real memory, power, and frame rate limits using tools like QNN, AIMET, TensorRT, and SNPE.

Work the low-level vision problems. Denoising, super-resolution, and frame interpolation are the core of the product, alongside depth, optical flow, and segmentation.

Ship into real deployments. The model goes out to partners including the Sony sensors division, Anduril, and the US Army and Air Force, where it either works in the field or it does not.