AI Ups the Resolution of Images

Entropix is an Oakland, California-based startup that uses deep learning to reconstruct video and still images to 9x its captured pixel density. Entropix calls this, “resolution on demand.”

The startup demonstrated its patented technology at the GPU Technology Conference in San Jose, California this week.

The technology has the potential to help law enforcement locate lost children and can be even implemented in autonomous vehicle technology.

“The higher the video resolution available on demand to the car’s object classification system, the more quickly a child in the road can be identified as such, and the more time and distance the car has to respond accordingly,” said Nathan Wheeler, Founder and Chief Product Officer at Entropix.

Wheeler and his team use NVIDIA Tesla V100 GPUs with the TensorRT inference accelerator to reconstruct (increase resolution) the images. ​

“We found a way to use the computational power of GPUs in order to get around the law of physics and be able to achieve resolution much higher and at much higher fidelity,” Wheeler said.

A model of Entropix’s technology

The startup says they work with different camera manufacturers to build and implement the dual sensor technology needed for the image reconstruction.

“More pixels can save lives,” Wheeler highlighted.

Read more>