Announcing an edge AI image processing model that can be converted to low noise images in real time (monoist) --Yahoo! News

delivery

Output image by LeapMind's minimal quantization model

　LeapMind announced on January 31, 2022 a "AI (Artificial Intelligence) Imaging Model" that can be operated on the edge device. It enables real -time movements and high quality image quality processing, which has been a challenge for edge devices so far, can be improved to a smartphone -like image quality. The evaluation version will be provided in February of the same year. [Original input image with a lot of noise] This model uses the LeapMind ultra-low power consumption AI reasoning accelerator IP "Efficiera" to reduce noise reduction models and deep learning base noise. Provides a learning model that can be processed and re -learned. In addition to the performance scalability of Efficiera, which can shoot real -time videos, the company's "Pixel Embedding Technology" and "extremely small quantumization technology" have enabled high performance, high -quality, and lighter weight. As a result, AI improves image quality, even without high -cost high -sensitivity sensors and large lenses, and provides image quality as well as the conventional 32 -bit floating point model. In addition, even video cameras used under low illuminance and low exposure conditions, such as surveillance cameras and inspection cameras, can be expected to improve object recognition and accuracy of tests.

Monoist