DeepNight is building night vision with AI software. We are pushing the state of the art of computational imaging to build a new generation of night vision goggles that see in the dark using AI. We conduct novel research in computational photography, novel efficient neural architectures, and model compression techniques to build extremely efficient AI on the edge.
We use PyTorch on 8x H100 GPUs. We deploy our night vision AI onto silicon on chips. Techniques like model compression, quantization aware training, knowledge distillation, are utilized. Efficient image processing algorithms are implemented in Halide and C++.