3D Depth Sensing on Neural Processors
Our Depth sensing solution is based on Deep Learning models
– It fuses in real-time an RGB image with ultra-low-resolution 3D ToF data to create a high-resolution depth map
– Achieved quality on par with Apple LiDAR-based TrueDepth
– Outperforms iToF-based and Google’s RGB-based depth maps estimations
Designed and optimized for Neural Processing Units
– 40ms per frame – 25 fps on Hexagon Processor
– 40ms per frame – 25 fps on Hexagon Processor
