Object classification and detection with deep learning in the box

Today, artificial intelligence often runs in the cloud, with data being sent back and forth between the application and the AI algorithms running on energy-guzzling cloud infrastructure. Easics uses its expertise in system-on-chip design to develop small, low-power and affordable AI engines that run locally, close to your sensors, e.g. inside a smart camera, sorting machine, robot or vehicle. The result is more secure, faster, and has a low and predictable latency.

easics deep learning in the box is based on convolutional neural networks. Since 2012 deep learning classification algoritms outperform classical vision techniques. The huge amount of data made available through cloud providers made it possible to train these networks. New architectures of these networks are constantly popping up and the hardware is so performant to calculate the convolutions in record time.  During training the network will define itself wich features it will use to classify an object. Deep learning is extremely effective to detect small variations in wood, textile, flowers,  ... and to detect anomalies and contamination in any kind of surface. 

easics’ embedded AI solutions integrate tightly with novel and existing sensors such as image sensors capturing light inside and outside the visible spectrum (such as hyperspectral and thermal infrared), 3D scanning laser (LiDAR), Time-of-Flight (ToF) sensors, radar, microscopy, ultrasound sensors, and microphones.

Deep learning object detection
 

Getting AI Out of the cloud

If you require to embed Artificial intelligence (based on a trained deep learning network) in your system, easics deep learning in the box is the solution for you. Via a standard ethernet or PCIe interface we can connect the AI engine inside your application. Where your application is running on a (existing) CPU and the AI engine is based on an FPGA. The integration with the application software is done via an API. The plug and play solution can make any device smart by deploying the neural network of your choice. Our current solution works with an Intel ARRIA 10 but Xilinx is also possible. If you would like to create your own baseboard for the AI engine we recommend you to use an FPGA SoM including memory with the pre-loaded easics deep learning core. Let's discuss how we can get your application out of the cloud by creating your AI engine.

 

AI engine
 

Embedded AI: features and applications

easics embedded AI demo deep learning

Why choose  easics deep learning in the box: 

  • Low and fixed latency 
  • Low power including battery powered applications 
  • Compact
  • Longevity  
  • Friendly hardware integration with a standard interface
    • TCP/IP or
    • PCIexpress
  • Flexible software stack and plug and play API integration
  • Scalable solution for future product roadmap 

Applications that can benefit from embedded AI can be found in industry 4.0, agriculture, healthcare, automotive and space: 

  • Quality control in manufacturing of semiconductors, electronics, textiles, wood, chemicals or pharmaceuticals
  • Application-specific machine vision systems
  • Smart cameras
  • Real time image classification for sorting of flowers, vegetables or fruits
  • Smart city: people, crowd and traffic monitoring
  • Smart health: medical image analysis, low-power wearables and implants
  • Smart mobility: real Time detection of objects for autonomous driving and robotics
 
 

Evaluation kit

We can provide you an ARIA 10 SoM development kit. The input and output interface is ethernet. We can provide this with the DNN of your choice e.g. Yolo V2 or V3, Resnet, mobilenet, tinyyolo, ... 

Deep learning FPGA development kit

If PCIe is your preferred interface another ARIA10 SoM carrier board is available with the DNN of your choice e.g. Yolo V2 or V3, Resnet, mobilenet, tinyyolo, ... 

Deep learning PCIe carrier board

Share:

Request a demo or evaluation kit!