Amol Borkar, Director of Product and Marketing for Vision and AI DSPs at Cadence Tensilica, demonstrates the company’s latest edge AI and vision technologies and products at the September 2023 Edge AI and Vision Alliance Forum. Specifically, Borkar demonstrates two applications running on customers’ SoCs, showcasing Cadence’s pervasiveness in AI.
The first demonstration is of people detection and tracking, performed on MediaTek’s Genio 500 SoC. As part of the demo, the camera captures an image and a person is detected using an objection detection AI network, followed by an OpenPose network which in realtime is able to overlay a skeletal representation of the person’s pose. The AI computation required for this demo is performed on MediaTek’s integrated APU (AI Processing Unit), which is comprised of 2x Tensilica Vision P6 DSPs. Additionally, a servo controlled articulating arm moves the camera to actively track the person’s movement.
The second demonstration is of a large image language model in an industrial camera, provided by Labforge.ca. Unlike typical LLM demos, this demo takes images and text prompts as input and provides a confidence level that the prompts match the captured image in realtime. The “Bottlenose” camera system features an SoC by Toshiba that integrates 4x Tensilica Vision P6 DSPs to help assist and accelerate the performance of today’s generative AI models. The Vision P6 DSP handles pre/post-processing, various CV/imaging algorithms, and multiple layers of the neural network.