LeapMind unveils AI image processing model for edge devices

Machine learning solutions provider LeapMind has unveiled a new artificial intelligence (AI)-powered image processing model capable of operating on edge devices. The new solution works through the company’s ultra-low power consumption AI inference accelerator IP Efficiera to reduce noise and improve the image quality of images in real-time.

Traditionally, this has been challenging as AI image processing required a high calculation cost, which was difficult to achieve on edge devices.

In addition, extremely low bit quantization for AI image processing has often caused quality degradation in resulting images.

The novel technology aims to solve these challenges of performance and image quality accuracy by making the low-bit quantization technology lighter, as well as combining it with the performance scalability of Efficiera v2.

“As far as we searched, we are the world’s first in bringing this image processing AI model into the product by low bit quantization technology,” comments LeapMind CTO Hiroyuki Tokunaga.

In terms of real-world applications, this may translate to industrial cameras having improved object recognition accuracy by enhancing the image quality even in low-light conditions.

Image quality is also a significant concern for accuracy in biometrics applications, particularly for photos taken by devices at the network edge.

“This model is a product that can be put into practical use only because of LeapMind focusing on both hardware and software development and we believe that it shows the new value of extremely low bit quantization,” Tokunaga adds.

According to LeapMind, the company’s Pixel embedding technology now offers image performance and quality comparable to 32-bit floating-point models.

This means that image quality can be improved by AI even on small-factor lenses such as those found in smartphones, thus removing the need for expensive high-sensitivity sensors and large lenses.

The new AI image processing model by LeapMind is scheduled to become available as an evaluation version at some point this month. The company has set up an email address for interested parties.

Article Topics

accuracy  |  AI  |  biometrics  |  biometrics at the edge  |  edge AI  |  edge cameras  |  machine learning  |  object recognition  |  research and development