ArcSoft and Cadence partner to develop AI and vision applications

ArcSoft has collaborated with Cadence to port beauty shot, high dynamic range (HDR), bokeh and facial unlock applications to the Vision P6 DSP. The joint solution has been designed into an applications processor from a leading global provider and is now shipping in smartphones.

According to Cadence, the lower power consumption on the Vision P6 DSP enables the bokeh effect to run in real time during live preview and capture. Facial unlock requires faster response time at lower power, and ArcSoft’s facial recognition algorithms are built for high efficiency, low power consumption and high accuracy.

ArcSoft says its HDR technology works in a variety of lighting conditions and intelligently adjusts and boosts brightness in underexposed areas while restoring details in overexposed areas. These applications lend themselves well to processing on the low-power, high-performance and highly programmable Vision P6 DSP, it adds.

The Vision P6 DSP has been designed into application processors from two of the top five mobile application processor vendors. Based on the Cadence Tensilica Xtensa architecture, the Vision P6 DSP combines flexible hardware options with an extensive library of vision/imaging DSP functions and numerous vision/imaging applications from established ecosystem partners.

The Vision P6 DSP supports AI applications developed in the Caffe, TensorFlow and TensorFlowLite frameworks through the Tensilica Xtensa Neural Network Compiler. The Vision P6 DSP also supports the Android Neural Network API for on-device AI acceleration in Android-powered devices.

The software environment includes complete, optimised support for more than 1,500 OpenCV-based vision and OpenVX 1.1 library functions. Rounding out the Tensilica Vision DSP family are the Vision Q6 DSP for vision and AI, which is suited for high-performance systems and applications with its 384GMAC/sec peak performance; the Vision C5 DSP, a complete, standalone DSP for neural networks that has 1TMAC/sec computational capacity; and the vision-optimised Vision P5 DSP.